VSD-HDP

This GitHub repository is created as part of attending the VLSI Hardware Development Program (VSD-HDP, Cohort: 21-Oct. 2023 to 29-Dec. 2023, transferred to 20-Jan. 2024 to 30-Mar. 2024 ).

Table of Contents (with Progress Status)

Day #	Topic(s) Covered	Status
Day 0	GitHub repo creation, System/ Tools Setup
Day 1	Introduction to Verilog RTL design and Synthesis
Day 2	Familiarization of .lib file structure and various timing models (QTMs/ETMs) Hierarchical vs. Flat synthesis Various Flip-Flop designs
Day 3	Logic Synthesis Optimizations Combinational Logic Optimizations Sequential Logic Optimizations
Day 4	Gate Level Simulation (GLS) and Synthesis-Simulation mismatch
Days 5-9	RISC-V based MYTH (Microprocessor for You in Thirty Hours) Day 5 Introduction to RISC-V ISA and GNU Compiler Toolchain Day 6 Introduction to ABI and basic verification flow Day 7 Digital Logic with TL-Verilog and Makerchip Day 8 Basic RISC-V CPU Microarchitecture Day 9 Complete Pipelined RISC-V CPU Microarchitecture
Day 10	GLS of the implemented RISC-V CPU Core Conversion of TL-Verilog code to Verilog/ SystemVerilog GLS of the RISC-V CPU Core
Days 11-13	Advanced Synthesis and STA with DC Day 11 Introduction to STA Day 12 Advanced SDC Constraints Day 13 STA using OpenSTA Pending: Documentation
Days 14-18	Circuit Design using SKY130 PDK Day 14 CMOS Fundamentals Day 15 Velocity Saturation and CMOS Inverter VTC Day 16 CMOS Switching Threshold and Dynamic Simulations Day 17 CMOS Noise Margin Robustness Evaluation Day 18 CMOS Power Supply and Device Variation Robustness Evaluation
Day 19	PVT Corner Analysis (Post-Synthesis Timing)
Days 20-24	Advanced Physical Design using OpenLANE/ Sky130 Day 20 Inception of open-source EDA, OpenLANE and Sky130 PDK Day 21 Floorplan considerations, Placement, Library Cells Day 22 Library Cell design using Magic and Characterization using Ngspice Day 23 Pre-layout timing analysis and importance of good clock tree Day 24 Final steps for RTL2GDS using tritonRoute and openSTA
Day 25	Post-placement STA analysis of your Design (RISC-V Myth CPU Core) * on ss,ff,tt corners * Compare with Post-synth
Day 26	Post-CTS, Post-Routing STA analysis of your Design * on ss,ff,tt corners * Compare with previous results
Day 27	Re-run the automated flow with ftarget of 100MHz and perform STA at various stages Pending: Documentation

Misc.	A collection of related miscellaneous information and notes.

Day 0: GitHub repo creation, System/ Tools Environment Setup

0.1 System Setup

The program uses Open Source EDA tools and thus a linux OS environment is preferred.

Virtual Machine Requirements:

• OS: Ubuntu 18.04+
• RAM: 6GB or above
• HDD: 50GB or above
• CPU cores: 4 vCPUs or more

0.2 Tools Setup

0.2.1 yosys – Yosys Open SYnthesis Suite

We will be building Yosys from the source using gcc, the instructions for which are available in the official README for yosys in its GitHub repo. Build Yosys from source

• Clone source code from git repository:

  git clone https://github.com/YosysHQ/yosys.git

• Install the build system prerequsites/ dependencies:

  sudo apt-get install build-essential clang bison flex \
      libreadline-dev gawk tcl-dev libffi-dev git \
      graphviz xdot pkg-config python3 libboost-system-dev \
      libboost-python-dev libboost-filesystem-dev zlib1g-dev

• Build (using gcc) and install Yosys:

  make config-gcc
  make
  sudo make install

• Invoke from shell:
  

0.2.2 iverilog - Icarus Verilog

• Install from official Ubuntu repository:

  sudo apt-get install iverilog

• Invoke from shell:
  

0.2.3 GTKWave

• Install from official Ubuntu repository:

  sudo apt-get install gtkwave

• Invoke from shell:
  

---

Day 1: Introduction to Verilog RTL design and Synthesis

1.1 Lab: Functional Simulation of RTL design using iverilog and gktwave

In this lab, we clone the gihthub repo - sky130RTLDesignAndSynthesisWorkshop - and use one of the example RTL design sources together with its corresponding testbench to get familiarised with the functional simulation of an RTL design using iverilog and gtkwave.

1. Clone the github repo with the RTL design examples and sky130*.lib files

git clone https://github.com/kunalg123/sky130RTLDesignAndSynthesisWorkshop.git

2. We will be performing functional simulation of the module good_mux (defined in good_mux.v), as an example.

# Step1: Use iverilog to read and interpret the source and testbench file(s) and generate 
# a compiled output (with default output format=vvp)
# Syntax: iverilog -o <outfile> <file1.v> <file2.v> ... <tb_top.v>

iverilog -o good_mux good_mux.v tb_good_mux.v


# Step 2: iverilog has now generated a compiled output named "good_mux", which can now be run using
# the simulation runtime engine, vvp
# Syntax: vvp <outfile>

vvp good_mux


# Step 3: Running the verilog simulation in step 2 will generate & dump the stimulus and output 
# signal values for the defined simulation duration in the testbench to a vcd file.
# This can now be viewed using gtkwave.
# Syntax: gtkwave <dumpfile.vcd>

gtkwave tb_good_mux.vcd

Snapshot of the waves from the above simulation in gtkwave:

1.2 Lab: Synthesis of RTL design using Yosys with sky130 library as target

In this lab, we will perform gate-level synthesis of the example RTL design simulated in the previous session using Yosys and sky130 as the target library.

# 1. Invoke the yosys shell (since we are doing each step manually this time around)
yosys

# 2. Read the library file(s)
# Syntax: read_liberty -lib <path-to-libfile(s)>

read_liberty -lib ../my_lib/lib/sky130_fd_sc_hd__tt_025C_1v80.lib


# 3. Read the RTL design files
# Syntax: read_verilog <verilog_file.v>

read_verilog good_mux.v


# 4. Perform the synthesis
# Syntax: synth -top <module_name> 

synth -top good_mux


# 5. Generate the gate-level netlist for the target library
# Syntax: abc -liberty <path-to-libfile(s)>
# Note: Pay attention to the syntesis results!

abc -liberty ../my_lib/lib/sky130_fd_sc_hd__tt_025C_1v80.lib

------------------------------------------------------
ABC Result:
------------------------------------------------------
4.1.2. Re-integrating ABC results.
ABC RESULTS:   sky130_fd_sc_hd__mux2_1 cells:        1
ABC RESULTS:        internal signals:        0
ABC RESULTS:           input signals:        3
ABC RESULTS:          output signals:        1
------------------------------------------------------


# 6. (Optional) Show a graphical flow diagram of the logic realized by the 
# synthesis tool for the provided target library file(s)
# Syntax: show
show


# 7. Write the output netlist
# Syntax: write_verilog <netlist_outfile.v>
#         write_verilog -noattr  <netlist_outfile.v>  (excluding attributes)

write_verilog -noattr good_mux_netlist.v

Generated Netlist:
-------------------------------------------------------------------------------------------
/* Generated by Yosys 0.34+43 (git sha1 d21c464ae, gcc 9.4.0-1ubuntu1~20.04.2 -fPIC -Os) */

module good_mux(i0, i1, sel, y);
  wire _0_;
  wire _1_;
  wire _2_;
  wire _3_;
  input i0;
  wire i0;
  input i1;
  wire i1;
  input sel;
  wire sel;
  output y;
  wire y;
  sky130_fd_sc_hd__mux2_1 _4_ (
    .A0(_0_),
    .A1(_1_),
    .S(_2_),
    .X(_3_)
  );
  assign _0_ = i0;
  assign _1_ = i1;
  assign _2_ = sel;
  assign y = _3_;
endmodule
-------------------------------------------------------------------------------------------

Logic realized by the synthesis tool in the above example:

---

Day 2: Timing libs, Hierarchical vs. Flat Synthesis, Flip-Flop coding styles

2.1 Familiarization of the .lib file structure and various timing models (QTMs/ETMs)

1. Open up the sky130_fd_sc_hd__tt_025C_1v80.lib library file and get familiarized with the overall structure and the various parameters defined like:

• PVT corner for which the lib file is defined
• Process technology
• Delay model
• For the different standard cells, various parameters like: Cell Area, Pin capacitances, Leakage power for different input combinations, Dynamic power during transitions, Transition times, Propagation Delay times etc.

2. Read up about the various timing models (QTM, ETM, NLDM etc.)
   Ref:
   [1]: STA for Nanometer Designs - J. Bhasker, Rakesh Chadha (Chapter 3)
   [2]: https://www.vlsi-expert.com/2011/02/etm-extracted-timing-models-basics.html
   [3]: https://asic-pd.blogspot.com/2011/08/basic-of-timing-analysis-in-physical_22.html

2.2 Lab: Hierarchical vs. Flat Synthesis

In hierarchical synthesis, the design is broken down into multiple hierarchical levels. Each level represents a different abstraction of the design, ranging from high-level functional blocks to detailed logic gates.
The ASIC or chip gets divided into smaller and simpler modules or blocks, each having its own functionality and interface, and the top-level module defines the connectivity of the various instantiation(s) of these sub-modules thus realizing the overall functionality/ behavior as well as performance of the chip.

• Advantages:

  • Modularity & reusability: Allows designers to manage complexity by breaking down the design into manageable blocks and thus allowing reusability of these blocks.
  • Ease of Design: Reduced complexity now that each team is dealing with smaller sub-modules. This facilitates a "Divide and conquer" approach in parallelising the design effort as different teams and/or engineers can work on different modules independently leading to improved productivity. Faster runtimes for synthesis and analysis EDA tools since each team will be working on individual modules which are small compared to the complete chip.
  • Optimization: Each block can be synthesized independently, optimizing performance and area at each level.

• Applications:

  • Commonly used in complex designs where the entire circuit cannot be synthesized at once due to size or complexity, like modern SoCs, CPUs, GPUs, DSPs etc.

• Disadvantages:

  • Hierarchical Partitioning Overhead: Creating and managing multiple hierarchical levels requires careful planning and overhead in terms of design partitioning and inter-level communication.
  • Timing Closure Challenges: Timing closure can be more challenging in hierarchical synthesis due to the interaction between timing constraints at different levels. Ensuring timing requirements are met across all levels can be complex.
  • Verification Complexity: Verification tasks become more complex with hierarchical designs, as each level may require separate verification efforts before integration testing can be performed.
  • Design Consistency: Maintaining design consistency across different levels of hierarchy (especially with different designers working on different modules) can be a challenge, leading to potential integration issues.

Perform the hierarchical synthesis from the Yosys shell using the following commands:

read_liberty -lib ../lib/sky130_fd_sc_hd__tt_025C_1v80.lib
read_verilog multiple_modules.v
synth -top multiple_modules
abc -liberty ../lib/sky130_fd_sc_hd__tt_025C_1v80.lib
write_verilog -noattr multiple_modules_hier.v
show -stretch multiple_modules
show sub_module1
show sub_module2

Synthesis result preserving the design hierarchy:

2.2.2 Flat Design

Flat design approach considers the entire ASIC/ chip as a single monolithic entity. There is no concept of sub-modules or hierarchy andthe entire design is synthesized as a single unit without hierarchical decomposition, directly into the standard cells, IPs and macros available in the design library.

• Advantages:

  • Simplicity: Easier for small designs where breaking into hierarchical levels might not be necessary.
  • Performance Optimization: Can potentially optimize performance across the entire design as a whole.

• Applications:

  • Suitable for smaller designs where the overhead of managing hierarchical levels may not be justified. Often used in simpler digital circuits or where the design size does not necessitate hierarchical partitioning.

• Disadvantages:

  • Scalability Issues: Flat synthesis becomes impractical for very large designs due to the sheer size and complexity, making it difficult to handle efficiently within a single synthesis pass.
  • Limited Modularity: Lack of hierarchical decomposition reduces modularity, making it harder to reuse or independently optimize different parts of the design.
  • Complexity in Optimization: Optimizing a large and complex design as a whole can be challenging, as changes in one part of the design can have unintended consequences elsewhere.
  • EDA Tool Efficiency: Synthesis tools may struggle to handle large flat designs efficiently, resulting in longer synthesis times or potential resource constraints.

To flatten the hierarchical design in Yosys, the command flatten is used following which we can write the netlist, as shown below:

flatten
write_verilog -noattr multiple_modules_flat.v
show -stretch multiple_modules 

Synthesis result after flattening:

Comparing the netlist generated for hierarchical and flat synthesis:

---

2.3 Lab: Various Flip-Flop Designs

Here, we will take a look at the simulation and synthesis of different flip-flops.

2.3.1 DFF with Asynchronous Reset

module dff_asyncres ( input clk ,  input async_reset , input d , output reg q );
always @ (posedge clk , posedge async_reset)
begin
    if(async_reset)
  	  q <= 1'b0;
    else	
  	  q <= d;
end
endmodule

NOTE : To synthesize flip-flops using Ysosys, we need to provide an additional command dfflibmap so as to map the internal flip-flop cells to the flip-flop cells in the technology library specified in the given liberty file.

# For example, to perform synthesis of the dff_asyncres module for the sky130 library:

read_liberty -lib ../lib/sky130_fd_sc_hd__tt_025C_1v80.lib 
read_verilog dff_asyncres.v 
synth -top dff_asyncres 
dfflibmap -liberty ../lib/sky130_fd_sc_hd__tt_025C_1v80.lib 
abc -liberty ../lib/sky130_fd_sc_hd__tt_025C_1v80.lib 
show

Behavioral Simulation:

Reset Deassertion	Reset Assertion

Synthesis Result:

2.3.2 DFF with Synchronous Reset

module dff_syncres ( input clk , input async_reset , input sync_reset , input d , output reg q );
always @ (posedge clk )
begin
    if (sync_reset)
  	  q <= 1'b0;
    else	
  	  q <= d;
end
endmodule

Behavioral Simulation:

Reset Deassertion	Reset Assertion

Synthesis Result:

2.3.3 DFF with both Asynchronous & Synchronous Reset

module dff_asyncres_syncres ( input clk , input async_reset , input sync_reset , input d , output reg q );
always @ (posedge clk , posedge async_reset)
begin
    if(async_reset)
  	  q <= 1'b0;
    else if (sync_reset)
  	  q <= 1'b0;
    else	
  	  q <= d;
end
endmodule

Behavioral Simulation:

Synthesis Result:

---

2.4: Some interesting synthesis optimizations involving multipliers

Here, we will take a look at the synthesis of two special cases of multipliers where no cells are used at all.

2.4.1 Multiply by 2

The input is an n-bit binary number and the output is twice the input.

i.e., output[n:0] = 2 * input[n-1 : 0]

This is equivalent to left-shifting the input by 1 bit position with trailing 0 added in the LSB.
In this case also, we can see that there aren't any cells needed to realize this logic,
and the input to output interconnections involve just wires.  

Netlist generated by Yosys:

module mul2(a, y);
  input [2:0] a;
  wire [2:0] a;
  output [3:0] y;
  wire [3:0] y;
  assign y = { a, 1'h0 };
endmodule

Synthesis result:

2.4.2 Multiply a 3-bit number by 9

The input is a 3-bit binary number and the output is defined to be 9 * input.

i.e., output[5:0] = 9 * input[2:0]

In other words,  
    output[5:0] = 4'b1001 * input[2:0]  
    output[5:0] = {input[2:0], input[2:0]}

In this case also, we can see that there aren't any cells needed to realize this logic,
and the input to output interconnections involve just wires.  

Netlist generated by Yosys:

module mult9(a, y);
  input [2:0] a;
  wire [2:0] a;
  output [5:0] y;
  wire [5:0] y;
  assign y = { a, a };
endmodule

Synthesis result:

---

Day 3: Logic Optimization

Logic optimization is a process of finding an equivalent representation of the specified logic circuit under one or more specified constraints. The goal of logic optimization of a given circuit is to obtain the smallest logic circuit that evaluates to the same values as the original one.

In terms of Boolean algebra, the optimization of a complex Boolean expression is a process of finding a simpler one, which would upon evaluation ultimately produce the same results as the original one.

3.1 Combinational Logic Optimizations

The combinational logic is simplified to the most optimized form which is efficient in terms of area & power savings.

3.1.1 Constant Propagation

This is a direct optimization method wherein the Boolean expression of the synthesized logic is simplified if any of the inputs are "a constant" and subsequently some of the logic gate outputs also propagate a constant value always.

3.1.2 Boolean Logic Optimization

The various Boolean expression optimization techniques like K-maps (graphical), Quine-McLusky, reduction to standard SOP/ POS forms best suited for the cell library/ technology etc.

Note:

• The command to perform logic optimization in Yosys is opt_clean.
• Additionally, for a hierarchical design involving multiple sub-modules, the design must be flattened by running the flatten command before executing the opt_clean command.

USAGE:
After the synth -top <module_name> command is executed, do:
    opt_clean -purge

This command identifies wires and cells that are unused and removes them.
The additional switch, purge also removes the internal nets if they have a public name.

# Example showing the sequence of commands to perform combinational logic optimization using Yosys
# on module opt_check in opt_check.v:
    read_liberty -lib ../lib/sky130_fd_sc_hd__tt_025C_1v80.lib 
    read_verilog opt_check.v 
    synth -top opt_check 
    opt_clean -purge
    abc -liberty ../lib/sky130_fd_sc_hd__tt_025C_1v80.lib 
    show

3.1.3 Lab 1: opt_check.v

module opt_check (input a , input b , output y);
    assign y = a?b:0;
endmodule

// We can see by direct simplification that:
//      y = a*b + a_bar*0 = ab
// i.e, a 2-input AND Gate. 

Synthesis Result:

---

3.1.3 Lab 2: opt_check2.v

module opt_check2 (input a , input b , output y);
    assign y = a?1:b;
endmodule

// Boolean simplification:
//      y = a + a_bar*b
//        = ab + a*b_bar + a_bar*b
//        = (ab + ab) + a*b_bar + a_bar*b
//        = (ab + a*b_bar) + (ab + a_bar*b)
//        = a + b 
// i.e, a 2-input OR Gate. 

Synthesis Result:
The cell library already seems to have an OR gate available as the synthesis result is an OR gate itself and not a NAND-realization of the OR to avoid the stacked PMOS as shown in the demo videos

---

3.1.3 Lab 3: opt_check3.v

module opt_check3 (input a , input b, input c , output y);
    assign y = a?(c?b:0):0;
endmodule

// Boolean simplification:
//      y = a* (bc) + a_bar*0
//        = abc
// i.e, a 3-input AND Gate.

Synthesis Result:

---

3.1.3 Lab 4: opt_check4.v

module opt_check4 (input a , input b , input c , output y);
    assign y = a?(b?(a & c):c):(!c);
endmodule

// Boolean simplification:
//      y = a*(abc + b_bar*c) + a_bar*c_bar
//        = abc + a*b_bar*c + a_bar*c_bar
//        = abc + a*b_bar*c + (a_bar*b*c_bar + a_bar*b_bar*c_bar)
//        = ac + a_bar*c_bar
//        = 
// i.e, a 2-input XNOR Gate.

Synthesis Result:

---

3.1.3 Lab 5: multiple_module_opt.v

module sub_module1(input a , input b , output y);
    assign y = a & b;
endmodule

module sub_module2(input a , input b , output y);
    assign y = a^b;
endmodule

module multiple_module_opt(input a , input b , input c , input d , output y);
    wire n1,n2,n3;

    sub_module1 U1 (.a(a) , .b(1'b1) , .y(n1));
    sub_module2 U2 (.a(n1), .b(1'b0) , .y(n2));
    sub_module2 U3 (.a(b), .b(d) , .y(n3));

    assign y = c | (b & n1); 
endmodule

Synthesis Result:

---

3.1.3 Lab 6: multiple_module_opt2.v

module sub_module(input a , input b , output y);
    assign y = a & b;
endmodule

module multiple_module_opt2(input a , input b , input c , input d , output y);
    wire n1,n2,n3;

    sub_module U1 (.a(a) , .b(1'b0) , .y(n1));
    sub_module U2 (.a(b), .b(c) , .y(n2));
    sub_module U3 (.a(n2), .b(d) , .y(n3));
    sub_module U4 (.a(n3), .b(n1) , .y(y));
endmodule

Synthesis Result:

---

3.2 Sequential Logic Optimizations

3.2.1 Constant Propagation

Optimization technique used when a constant value is propagated through a flip-flop -- i.e., irrespective of the state of the triggering signals (CLK, Reset, Set), there are no transitions in the flip-flop output.

[Other Advanced optimization methods not covered by examples in detail:]

3.2.2 State optimization

Unused states in the sequential design are optimized and/or the total states needed in the FSM are minimized.

3.2.3 Cloning

This is a physically-aware (PnR-aware) optimization method where some of the flops in the design are cloned/ duplicated so that the timing can be met post-PnR for the timing arcs involved (provided there is already some minimum positive slack available).

3.2.4. Retming

The pipelining flops in the design are placed optimally so that the combinational delay at each pipeline stage is more or less equalized so that the maximum clock frequency can be increased.

# Example showing the sequence of commands to perform combinational logic optimization using Yosys
# on module dff_const1 in dff_const1.v:
    read_liberty -lib ../lib/sky130_fd_sc_hd__tt_025C_1v80.lib 
    read_verilog dff_const1.v 
    synth -top dff_const1
    opt_clean -purge
    dfflibmap -liberty ../lib/sky130_fd_sc_hd__tt_025C_1v80.lib 
    abc -liberty ../lib/sky130_fd_sc_hd__tt_025C_1v80.lib 
    show

3.2.5 Lab 1: dff_const1.v

module dff_const1(input clk, input reset, output reg q);
    always @(posedge clk, posedge reset)
    begin
	    if(reset)
                q <= 1'b0;
	    else
                q <= 1'b1;
    end
endmodule

In this example, no optimization is possible as the flop output, q changes.

Behavioral Simulation
Synthesis Result

---

3.2.5 Lab 2: dff_const2.v

module dff_const2(input clk, input reset, output reg q);
    always @(posedge clk, posedge reset)
    begin
    if(reset)
        q <= 1'b1;
    else
        q <= 1'b1;
    end
endmodule

Here, the the flip-flop output, q remains constant at 1 irrespective of the other signals in the sensitivity list.
Thus the realization does not need any cells and q is connected to 1'b1.

Behavioral Simulation
Synthesis Result

---

3.2.5 Lab 3: dff_const3.v

module dff_const3(input clk, input reset, output reg q);
    reg q1;

    always @(posedge clk, posedge reset)
    begin
        if(reset)
            begin
	         q <= 1'b1;
                q1 <= 1'b0;
            end
        else
	     begin
                 q1 <= 1'b1;
	          q <= q1;
	     end
    end
endmodule

Here, both q1 & q have transitions and thus cannot be optimized further.
So the design will have two DFFs.

Behavioral Simulation
Synthesis Result

---

3.2.5 Lab 4: dff_const4.v

module dff_const4(input clk, input reset, output reg q);
    reg q1;
    
    always @(posedge clk, posedge reset)
    begin
        if(reset)
            begin
                q <= 1'b1;
                q1 <= 1'b1;
            end
        else
            begin
                q1 <= 1'b1;
                q <= q1;
            end
        end
endmodule

In this example, both q1 & q remain constant at 1'b1.
Thus, the design can be optimized to have only wires. Further, q1 being an internal net can also be removed.

Behavioral Simulation
Synthesis Result

---

3.2.5 Lab 5: dff_const5.v

module dff_const5(input clk, input reset, output reg q);
    reg q1;
    
    always @(posedge clk, posedge reset)
    begin
        if(reset)
            begin
                q <= 1'b0;
                q1 <= 1'b0;
            end
        else
            begin
                q1 <= 1'b1;
                q <= q1;
            end
        end
endmodule

Behavioral Simulation
Synthesis Result

---

3.2.5 Lab 6: counter_opt.v

module counter_opt (input clk , input reset , output q);
    reg [2:0] count;
    assign q = count[0];
    
    always @(posedge clk ,posedge reset)
    begin
        if(reset)
            count <= 3'b000;
        else
            count <= count + 1;
    end
endmodule

This design serves as an example for Sequential logic optimization with designs having unused outputs.
Although we have a 3-bit up counter in the RTL design, only the LSB, count[0:0] is used for generating the output signal, q.
Since count[0:0] toggles every clock cycle, there really is a need for only one flip-flop in the circuit.
In other words, the synthesis output does not have a 3-bit up counter and its associated count incrementing logic.

Behavioral Simulation
Synthesis Result w/o opt_clean switch
Synthesis Result with opt_clean switch

---

3.2.5 Lab 7: counter_opt2.v

module counter_opt (input clk , input reset , output q);
    reg [2:0] count;
    assign q = (count[2:0] == 3'b100);

    always @(posedge clk ,posedge reset)
    begin
        if(reset)
            count <= 3'b000;
        else
            count <= count + 1;
        end
endmodule

In this design, we have a 3-bit up counter and all the bits in the counter state value, count[2:0] are used for generating the output signal, q.
q = 1, when count[2:0] = 3'b100
So when this design is synthesized, we expect 3 DFF instantiations to be present along with the count incrementing logic and the logic to generate, q.

Behavioral Simulation
Synthesis Result

---

Day 4: Gate Level Simulation (GLS) and Synthesis-Simulation mismatch

4.1 Gate Level Simulation (GLS)

• Gate Level refers to the netlist view of a design after the synthesis has been performed.
• RTL simulations are pre-synthesis, while GLS is post-synthesis - i.e., in RTL simulations, the Device Under Test (DUT) is the RTL design itself while in GLS, the DUT is the netlist generated after synthesis.
• The RTL code and the generated netlist are logically equivalent (well, supposed to be!) and hence the same testbenches can be used to verify both.
• Although it is expected that the generated netlist has the same logical correctness as the RTL design, there can sometimes be mismatches between the RTL-level simulation and the sythesized design (Synthesis - Simulation mismatch) and thus arises the need to run GLS to help identify such scenarios and fix them to ensure the logical correctness post-synthesis as well.

To run GLS, we need to provide the Gate level netlist, the testbench and the Gate Level verilog models to the simulator.
GLS can be run in different delay modes:

1. Functional validation (zero delay similar to RTL sim): if the Gate Level verilog models do not have the timing information for various corners, we can only verify the functional correctness of the design by running GLS.
2. Full Timing validation: if the Gate level verilog models have the necessary timing information, both the functional correctness and the timing behaviour can be verified by GLS.

4.1.1 GLS using iverilog

The following block diagram shows the GLS flow using iverilog:

The Gate level verilog model(s) need to be provided as shown below to do GLS using iverilog:

Syntax:
    iverilog <path-to-gate-level-verilog-model(s)> <netlist_file.v> <tb_top.v>

Example using ternary_operator_mux_netlist.v:
    iverilog ../my_lib/verilog_model/primitives.v ../my_lib/verilog_model/sky130_fd_sc_hd.v ternary_operator_mux_netlist.v tb_ternary_operator_mux.v

4.2 Synthesis - Simulation mismatch

Some of the common reasons for Synthesis - Simulation mismatch (mismatch between pre- and post-synthesis simulations) :

• Incomplete sensitivity list
• Use of blocking assignments inside always block vs. non-blocking assignments
  • Blocking assignments ("=") inside always block are executed sequentially by the simulator.
  • The RHS of non-blocking assignments ("<=") are evaluated first and then assigned to the LHS at the same simulation clock tick by the simulator.
  • Synthesis will yield the same circuit with blocking and non-blocking assignments, with the synthesis output being that of the non-blocking case for both.
  • Hence, if the RTL was written assuming one functionality using blocking assignments, a simulation mismatch can occur in GLS.
• Non-standard verilog coding

4.2.1 Lab: Example 1: ternary_operator_mux.v

module ternary_operator_mux (input i0 , input i1 , input sel , output y);
    assign y = sel ? i1 : i0;
endmodule

RTL Simulation
GLS
Synthesis Result

---

4.2.2 Lab: Example 2: bad_mux.v

module bad_mux (input i0 , input i1 , input sel , output reg y);
    always @ (sel)
    begin
        if(sel)
            y <= i1;
        else
            y <= i0;
    end
endmodule

RTL Simulation
GLS
Synthesis Result

In this case, we can clearly see that there is a mismatch in the simulation between pre and post-synthesis.
The pre-synthesis simulation shows a behavior resembling that of a posedge triggered DFF with the "sel" signal acting as the CLK and the "i1" acting as the D input.
The synthesis result, however, is a 2-input mux and not a DFF.
In fact, yosys actually throws a warning message about the possible omission of signals from the sensitivity list assuming a purely combinational circuit.

Yosys warning about missing signals in sensitivity list

---

4.2.3 Lab: Example 3: blocking_caveat.v

module blocking_caveat (input a , input b , input  c, output reg d);
    reg x;

    always @ (*)
    begin
        d = x & c;
        x = a | b;
    end
endmodule

In this case, there is a mismatch in the simulation results between pre and post-synthesis due to the use of blocking assignments. Assuming we wanted to implement just a combinational logic with output, d = (a + b) * c:

• In the RTL sim, the blocking assignments make it seem as if there is a flop in the design.
• While in the GLS, the design is synthesized to a O2A1 gate implementing d = (a + b) * c, with no flops inferred, thus resulting in the mismatch.

RTL Simulation
GLS
Synthesis Result

---

Days 5-9: RISC-V based MYTH (Microprocessor for You in Thirty Hours)

Day 5: Introduction to RISC-V ISA and GNU Compiler Toolchain

RISC-V is an open standard instruction set architecture based on established reduced instruction set computer(RISC) principles. It was first started by Prof. Krste Asanović and graduate students Yunsup Lee and Andrew Waterman in May 2010 as part of the Parallel Computing Laboratory, at UC Berkeley. Unlike most other ISA designs, the RISC-V ISA is provided under open source licenses that do not require fees to use, which provides it a huge edge over other commercially available ISAs. It is a simple, stable, small standard base ISA with extensible ISA support, that has been redefining the flexibility, scalability, extensibility, and modularity of chip designs.

5.1 RISC-V ISA base and extensions

RISC-V has a modular design, consisting of alternative base parts, with added optional extensions. The ISA base and its extensions are developed in a collective effort between industry, the research community and educational institutions. The base specifies instructions (and their encoding), control flow, registers (and their sizes), memory and addressing, logic (i.e., integer) manipulation, and ancillaries. The base alone can implement a simplified general-purpose computer, with full software support, including a general-purpose compiler.

The RISC-V ISA is defined as a Base integer ISA, which is the basic necessity for the implemetation of any CPU core. In addition to that it also has optional extensions to the base ISA. The base RISC-V ISA has a little-endian memory system. The standard is maintained by the RISC-V foundation.

The standard extensions are specified to work with all of the standard bases, and with each other without conflict.

5.2 RISC-V Instruction Formats

References:

1. https://en.wikipedia.org/wiki/RISC-V
2. RISC-V Technical Specifications
3. RISC-V MYTH Workshop

5.3 RISC-V ISA Simulator and Compiler toolchain setup

• The RISC-V ISA simulator & GNU Compiler toolchain can be installed by running the following script from the terminal: run.sh

  • Note:
    • We are not building the RISC-V C/C++ cross-compiler from source and instead taking a pre-built compiler from SiFive: https://static.dev.sifive.com/dev-tools/riscv64-unknown-elf-gcc-8.3.0-2019.08.0-x86_64-linux-ubuntu14.tar.gz.
    • To build the RISC-V GNU Compiler Toolchain from scratch, please refer to the steps in the following GitHub repo: https://github.com/riscv-collab/riscv-gnu-toolchain.
  
  

• The following GitHub repo - RISC-V Workshop Collaterals - contains sample programs and the verilog code for a RV32I processor core (PicoRV32).
  

• More information on how to use the toolchain to compile a source file, simulation of the object file and using the interactive debug mode can be found in the README page of the Spike RISC-V simulator GitHub repository: Spike RISC-V ISA Simulator
  

• More information on the RISC-V specific compiler options like -march, -mabi, and -mtune can be found in the following links:

  1. https://gcc.gnu.org/onlinedocs/gcc/RISC-V-Options.html
  2. https://www.sifive.com/blog/all-aboard-part-1-compiler-args
  3. https://five-embeddev.com/toolchain/2019/06/26/gcc-targets/
  4. https://github.com/riscv-non-isa/riscv-toolchain-conventions

5.3.1 Lab: RISC-V program simulation using Spike RISC-V ISA Simulator

• Spike is a RISC-V ISA Simulator that implements a functional model of one or more RISC-V harts. (A hart is a hardware thread in RISC-V terminology)

• pk is the RISC-V proxy kernel. It is a lightweight application execution environment that can host statically-linked RISC-V ELF binaries. It is designed to support tethered RISC-V implementations with limited I/O capability and thus handles I/O-related system calls by proxying them to a host computer.

• bbl, the Berkeley Boot Loader is a supervisor execution environment for tethered RISC-V systems. It is designed to host the RISC-V Linux port.

The following slide from the RISC-V Software Tools Bootcamp presentation shows the overview of the Spike + pk workflow:

Ref: Software Tools Bootcamp

Write a C program to compute the sum of first N natural numbers, compile using RISC-V GCC, simulate using Spike RISC-V ISA Simulator and disassemble to view the assembly code

C Program:

#include <stdio.h>

int main() {
    int n=100, i, sum = 0;
    for (i = 1; i <= n; ++i) {
        sum += i;
    }
    printf("Sum of first %d natural numbers = %d\n", n, sum);
    return 0;
}

Compilation command:

riscv64-unknown-elf-gcc -O1 -mabi=lp64 -march=rv64i -o sum1toN.o sum1toN.c
    where,
        -mabi=ABI-string option specifies the ABI (Application Binary Interface) to be used.
        -march=ISA-string option specifies the RISC-V ISA for which the object code is to be generated.
        -O<number>, -Ofast, -Os, -Og etc. specify the optimize option to be used by the compiler.

Spike simulation atop the proxy kernel (pk):

spike pk sum1toN.o

Output:

Disassemble command:

riscv64-unknown-elf-objdump -d sum1toN.o
    where,
        -d, --disassemble flag displays the assembler contents of the executable sections.

Output:

Comparing the generated assembly code for main function with -O1 vs. -Ofast compiler options:

-O1	-Ofast

5.3.2 Lab: Interactive Debug mode in Spike RISC-V ISA Simulator

Use the Interactive Debug mode in Spike RISC-V ISA sim to observe the execution of the program with -Ofast flag

A small example of how to use the debug mode can be found in the following page: Interactive Debug Mode

To enter the interactive debug mode, launch spike with -d option:

Example:
spike -d pk sum1toN.o

Snapshot showing usage of Spike Interactive Debug Mode

Interactive commands:
reg <core> [reg]                # Display [reg] (all if omitted) in <core>
freg <core> <reg>               # Display float <reg> in <core> as hex
fregh <core> <reg>              # Display half precision <reg> in <core>
fregs <core> <reg>              # Display single precision <reg> in <core>
fregd <core> <reg>              # Display double precision <reg> in <core>
vreg <core> [reg]               # Display vector [reg] (all if omitted) in <core>
pc <core>                       # Show current PC in <core>
priv <core>                     # Show current privilege level in <core>
mem [core] <hex addr>           # Show contents of virtual memory <hex addr>
                                  in [core] (physical memory <hex addr> if omitted)
str [core] <hex addr>           # Show NUL-terminated C string at virtual address <hex addr>
                                  in [core] (physical address <hex addr> if omitted)
dump                            # Dump physical memory to binary files
mtime                           # Show mtime
mtimecmp <core>                 # Show mtimecmp for <core>
until reg <core> <reg> <val>    # Stop when <reg> in <core> hits <val>
untiln reg <core> <reg> <val>   # Run noisy and stop when <reg> in <core> hits <val>
until pc <core> <val>           # Stop when PC in <core> hits <val>
untiln pc <core> <val>          # Run noisy and stop when PC in <core> hits <val>
until mem [core] <addr> <val>   # Stop when virtual memory <addr>
                                  in [core] (physical address <addr> if omitted) becomes <val>
untiln mem [core] <addr> <val>  # Run noisy and stop when virtual memory <addr>
                                  in [core] (physical address <addr> if omitted) becomes <val>
while reg <core> <reg> <val>    # Run while <reg> in <core> is <val>
while pc <core> <val>           # Run while PC in <core> is <val>
while mem [core] <addr> <val>   # Run while virtual memory <addr>
                                  in [core] (physical memory <addr> if omitted) is <val>
run [count]                     # Resume noisy execution (until CTRL+C, or [count] insns)
r [count]                         Alias for run
rs [count]                      # Resume silent execution (until CTRL+C, or [count] insns)
quit                            # End the simulation
q                                 Alias for quit
help                            # This screen!
h                                 Alias for help
Note: Hitting enter is the same as: run 1

---

Day 6: Introduction to ABI and basic verification flow

In computer software, an application binary interface (ABI) is an interface between two binary program modules. Often, one of these modules is a library or operating system facility, and the other is a program that is being run by a user. An ABI defines how data structures or computational routines are accessed in machine code, which is a low-level, hardware-dependent format.
Reference: Application binary interface

RISC-V CPU architecture has 32 registers. Application programmer, can access each of these 32 registers through its ABI name; for example, if we need to move the stack pointer, the command addi sp, sp, -16 will decrement the SP by 0x10, where "sp" is the ABI name of stack pointer. The following table shows the ABI Integer register calling convention :

For more detailed information, refer to the RISC-V ABI Specification v1.0

6.1 Lab: Rewrite the program to find the sum of first N natural numbers utilizing ABI function calls

C Program: 1to9_custom.c

#include <stdio.h>

extern int load(int x, int y); 

int main() {
    int result = 0;
    int count = 100;
    result = load(0x0, count+1);
    printf("Sum of number from 1 to %d is %d\n", count, result); 
}

Assembly program: load.s

.section .text
.global load
.type load, @function

load:
    add  a4, a0, zero //Initialize sum register a4 with 0x0
    add  a2, a0, a1   // store count of 10 in register a2. Register a1 is loaded with 0xa
                       // (decimal 10) from main program
    add	 a3, a0, zero   // initialize intermediate sum register a3 by 0
loop:
    add  a4, a3, a4   // Incremental addition
    addi a3, a3, 1    // Increment intermediate register by 1	
    blt  a3, a2, loop // If a3 is less than a2, branch to label named <loop>
    add  a0, a4, zero   // Store final result to register a0 so that it can be read by
                       // main program
    ret

The program can be compiled using the gcc and simulated using Spike as follows:

Disassembly of object code of above progam:

6.2 Lab: Simulate the above C program on a RISC-V CPU

For this exercise, we will use the design files from the following GitHub repository: https://github.com/kunalg123/riscv_workshop_collaterals.git

The following figure shows the Hardware + Firmware verification flow:

Basically what we are doing here is:

• Our objective is to use run the program as a testbench on the PicoRV32 RISC-V CPU processor design coded in the picorv32.v verilog file.
• For this, all the source files used are compiled into object code using gcc with rv32im as the target RISC-V ISA. Source file(s) include:
  • File(s) to perform the required CPU initialization
  • File(s) that define the various system calls used
  • File(s) containing the actual user program
• Then, the various object files are linked together into an ELF file using gcc
• The ELF file is then converted into a hex verilog memory file, firmware.hex using riscv64-unknown-elf-objcopy
• These 8-bit hex files are converted into 32-bit format using a python script, hex8tohex32.py
• Finally, a functional simulation is launched using iverilog, where the testbench loads the firmware.hex into the PicoRV32 CPU core for execution

To run the flow, execute the following file from shell:

rv32im.sh:

riscv64-unknown-elf-gcc -c -mabi=ilp32 -march=rv32im -o 1to9_custom.o 1to9_custom.c 
riscv64-unknown-elf-gcc -c -mabi=ilp32 -march=rv32im -o load.o load.S

riscv64-unknown-elf-gcc -c -mabi=ilp32 -march=rv32im -o syscalls.o syscalls.c
riscv64-unknown-elf-gcc -mabi=ilp32 -march=rv32im -Wl,--gc-sections -o firmware.elf load.o 1to9_custom.o syscalls.o -T riscv.ld -lstdc++
chmod -x firmware.elf
riscv64-unknown-elf-gcc -mabi=ilp32 -march=rv32im -nostdlib -o start.elf start.S -T start.ld -lstdc++
chmod -x start.elf
riscv64-unknown-elf-objcopy -O verilog start.elf start.tmp
riscv64-unknown-elf-objcopy -O verilog firmware.elf firmware.tmp
cat start.tmp firmware.tmp > firmware.hex
python3 hex8tohex32.py firmware.hex > firmware32.hex
rm -f start.tmp firmware.tmp
iverilog -o testbench.vvp testbench.v picorv32.v
chmod -x testbench.vvp
vvp -N testbench.vvp

---

6.3 Addendum: To view the waves in Gktwave

To dump the VCD file from the simulation using iverilog and further view the waves in Gktwave:

• In testbench.v, comment out the undef line for WRITE_VCD and add a define for the same:

  //`undef WRITE_VCD
  `define WRITE_VCD
  

• Additionally, in picorv32.v, add the following define to enable the debug wires to view the internal Register File:

  `define DEBUGREGS
  

Output:

• Modified source files:
  1to9_custom.c

  #include <stdio.h>
  
  extern int load(int x, int y); 
  
  int main() {
    int result = 0;
    int count = 10;
    result = load(0x0, count+1);
    printf("Sum of number from 1 to %d is %d\n", count, result); 
  }

  load.S

  .section .text
  .global load
  .type load, @function
  
  load:
      add  a4, a0, zero  // Initialize sum register a4 with 0x0
      add  a2, a0, a1    // storecount of 10 in register a2. Register a1 is loaded with 0xa (decimal 10) from main program
      add  a3, a0, zero  // Initialize intermediate sum register a3 by 0
  loop:
      add  a4, a3, a4    // Incremental addition
      addi a3, a3, 1     // Increment intermediate register by 1	
      blt  a3, a2, loop  // If a3 is less than a2, branch to label named <loop>
      add  a0, a4, zero  // Store final result to register a0 so that it can be read by main program
      ret

---

Day 7: Digital Logic with TL-Verilog and Makerchip

7.1 TL-Verilog

Transaction-Level Verilog (TL-Verilog) is an extension to SystemVerilog that supports a new design methodology, called transaction-level design. A transaction, in this methodology, is an entity that moves through structures like pipelines, arbiters, and queues. A transaction might be a machine instruction, a flit of a packet, or a memory read/write. Transaction logic, like packet header decode or instruction execution, that operates on the transaction can be placed anywhere along the transaction’s flow. Tools produce the logic to carry signals through their flows to stitch the transaction logic.

7.2 Makerchip Platform

Makerchip is a free online environment provided by Redwood EDA for developing integrated circuits. The online platform can be used to code, compile, simulate and debug Verilog designs from a web browser. In addition to Verilog and synthesizable SystemVerilog, the platform provides support for Transaction-Level Verilog (TL-Verilog).

The Makerchip IDE also provides various tutorials and examples (inside Learn menu in the main window) to help learn the new concepts and constructs in TL-Verilog quickly.

References:

1. https://www.redwoodeda.com/tl-verilog
2. https://www.tl-x.org
3. Top-Down Transaction-Level Design with TL-Verilog
4. Makerchip IDE
5. Makerchip IDE Documentation

In this session, we will learn the concepts and syntax of TL-Verilog by implementing basic combinational and sequential logic using the same in Makerchip.

1. Combinational logic
2. Sequential logic
3. Pipelined logic
4. Validity

7.2.1 Combinational Logic

Implemented basic combinational logic elements and circuits like gates, muxes, vector adder, combinational calculator etc. using TL-Verilog in Makerchip.

1. AND2 Gate

2. Vector Mux

3. Adder

4. Combinational Calculator

7.2.2 Sequential Logic

1. Fibonacci Series

2. Free-running Counter (starts from 1 and increments by 1 every cycle)

3. Sequential Calculator

7.2.3 Pipelined Logic

1. Pythagorus Theorem with 3-stage pipeline

2. Example showing the ease of pipelining/ retiming in TL-Verilog

3. Sequential Calculator with counter

4. 2-Cycle Sequential Calculator

7.2.4 Validity

Using validity makes the design cleaner. Debugging and error checking also becomes easier.

1. 2-Cycle Calculator with Validity
2. Calculator with Single-value Memory

---

Day 8: Basic RISC-V CPU Microarchitecture

Our objective is to implement a basic RISC-V CPU core (RV32I Base Instruction set except the instructions - FENCE, ECALL & EBREAK).
The following diagrams show the general block diagram of the CPU and the initial implementation pipeline diagram using TL-Verilog:

CPU Block Diagram
TL-Verilog based implementation pipeline/ flow diagram

The implementation is done stage-by-stage verifying the functionality at each step in the Makerchip IDE. The following logic blocks are implemented and verified:

• Program Counter (PC)
• Instruction Fetch from Instruction Memory (IMEM Read)
• Instruction Decoder
  • Instruction Type Decode (I, R, S, B, U, J)
  • Instruction Immediate Value Decoding
  • Instruction Field Decoding
  • Complete Instruction Decoding
• Register File Read
• Arithmetic & Logic Unit (ALU)
• Register File Write
• Branch Instructions
  • (Modifying PC logic and adding additional logic to handle the Branch instructions)

8.1 Program Counter + Instruction Fetch

Program Counter + Intruction Fetch

8.2 Instruction Decoder

Instruction Decoder
RISC-V Opcode Map
RISC-V Instruction Format
RV32I ISA Encoding
Instruction Type Decoding
Immediate Value Decoding

8.3 Register File Read

Register File Read
Register File module interface

8.4 ALU

ALU (only ADD, ADDI implemented for now)

8.5 Register File Write

Register File Write

8.6 Branch Instruction Logic

Branch Instruction Logic added

---

Day 9: Complete Pipelined RISC-V CPU Microarchitecture

Our RISC-V core from the previous day is still incomplete w.r.t the instructions implemented, and additionally we need to do pipelining and handling of the pipeline hazards.
We need to do the following to complete the CPU Design:

1. Pipeline the CPU, taking care of the data dependency & control flow hazards
2. Complete the implementation of the remaining ALU instructions
3. Implement DMEM & Load, Store instructions
4. Implement the Unconditional Jump (JAL, JALR) instructions

Pipelining the RISC-V CPU Core

9.1 Pipelining the CPU: Using 3-Cycle $valid signal

First, we will implement with a simplified 3-stage pipeline with using a 3-Cycle valid signal, the various stages being:

• PC
• Instruction Fetch + Decode
• RF Read, ALU
• RF Write, Branch Instrn. logic

This implementation would have an IPC of only ~1/3 as the valid signal is active once every 3 cycles (HLLHLL...) indicating only one valid instruction in the pipe at any point. We do this step to partition the core (or logic) into the respective pipeline stages first without having to worry about handling the pipeline hazards.

Waterfall Logic Diagram with 3-Cycle Valid
TL-V Logic Implementation Diagram

Makerchip-generated Block Diagram for 3-Cycle Valid design

9.2 Pipelining the CPU: Solving the data & control hazards

9.2.1 Data Hazard: Read-After-Write (RAW) in the Refister File

• There is a 2-cycle delay (by design) between RF Read and Write operations.
• Hence we have a Read-After-Write (RAW) data hazard if the current instruction in the pipe is trying to read from the Register File (RF) when the previous instruction had written to the same RF index.
• To solve this, we need to add a Register File Bypass Mux at the input of the ALU and select the previous ALU output if the previous instruction was writing to the RF index accessed in the current instruction.

Register File Bypass Waterfall Logic Diagram
Register File Bypass TL-V Implementation

9.2.2 Control Hazard: Branch Instructions

• We have control flow hazards when a branch is taken.
• The PC logic is updated to handle the case when a branch is taken or not.

Branch Instruction Control Hazard

9.3 Complete the ALU

The Instruction Decoder is updated to decode all the instructions and the complete ALU is implemented. Note: All load instructions are treated as the same as the LW instruction.

9.4 DMEM & Load, Store Instructions

9.4.1 DMEM

• The DMEM is a single-port R/W memory with 16 entries, 32-bit wide.
• The DMEM is placed in the 4th pipeline stage.

  DMEM

9.4.2 LOAD (LW, LH, LB, LHU, LBU) Instructions

• LOAD rd, imm(rs1)
• Loads the data from the DMEM address given by (rs1 + imm) to destination register provided by rd. i.e., rd <= DMEM(rs1 + imm)
  

9.4.3 STORE (SW, SH, LB) Instructions

• STORE rs2, imm(rs1)
• Stores the data from rs2 to the DMEM address given by (rs1 + imm). i.e., rd <= DMEM(rs1 + imm)
  

The $dmem_addr[3:0] is generated by the ALU by treating the load and store instructions to be equivalent to the ADDI instruction.

i.e., The ALU performs the following:
LOAD/ STORE : ($is_load || $is_s_instr) ? ($src1_value + $imm) 
ADDI        :                 $is_addi  ? ($src1_value + $imm) :

Since the DMEM is 32-bit wide and not byte or half-addressable:
$dmem_addr[3:0] = $result[5:2];

Muxes need to be placed at the inputs of RF write index ($rf_wr_index) and RF write data ($rf_wr_data) ports to select the appropriate values depending on the validity of the load instruction.

DMEM Load/ Store

Additionally, the Program Counter logic has to be updated for load redirects.

9.5 Unconditional Jump (JAL, JALR) Instructions

• JAL : Jump to (PC + IMM), equivalent to an unconditional branch w.r.t the calculation of the target address.
• JALR: Jump to (SRC1 + IMM)

The logic to calculate the branch target for JALR needs to be implemented.
The Program Counter logic also needs to be modified to handle the jumps.

9.6 Complete Pipelined RISC-V CPU Core Implementation in Makerchip

Click on the image below to open up the interactive svg file:

---

9.7 Bug found with the LW instruction and RF Read Bypass

Original Code: riscv_pipelined_with_LW_Bug.tlv

In the functional simulation of the RTL code in MakerChip IDE of the RISC-V CPU core that we have designed following the steps in the lecture videos and slides, I noticed two issues:

9.7.1 Issue #1

During the execution of the LW instruction, the DMEM address gets written to destination register in the first cycle.

(NOTE: This is a benign issue and not a concern)

• Since LW is an I-type (Immediate-type instruction), the $rd (Destination Register) is valid during this phase and thus $rf_wr_en (Register File Write Enable).

  // Immediate
  $is_i_instr = ($instr[6:2] == 5'b00000) ||
                ($instr[6:2] == 5'b00001) ||
                ($instr[6:2] == 5'b00100) ||
                ($instr[6:2] == 5'b00110) ||
                ($instr[6:2] == 5'b11001);
  ...
  $is_load  = ($opcode == 7'b0000011);
  ...
  $rd_valid = $is_r_instr | $is_i_instr | $is_u_instr | $is_j_instr;
  ...
  $rf_wr_en =  ($rd_valid && $valid && $rd != 5'b0) || >>2$valid_load;
  

• If we take the following example: m4_asm(LW, r15, r0, 00100).
  This instruction is supposed to do just: r15 [31:0] <= DMEM [(r0 + 00100)] [31:0]

  Due to our design, the DMEM address is generated by the ALU as: DMEM_addr = (rs1 + imm). Hence the ALU output (or the DMEM address) gets written to the destination register first and then two cycles later the actual data from the DMEM address gets written to the destination register.

• In our implementation, since it takes two cycles for valid data to be fetched from the DMEM and to be written to the destination register, we are squashing the 2 instructions already in the pipe in the "shadow" of the Load instruction.
  Hence writing this intermediate value to the destination register is not a concern a.
  Nevertheless, to avoid this unnecessary RF write for a cleaner implmentation, we can deassert $rf_wr_en for these two cycles for a valid load instruction.

  $rf_wr_en = (!$valid_load && !>>1$valid_load) && ($rd_valid && ($rd != 5'b0) && $valid) || >>2$valid_load;
  

9.7.2 Issue #2

The instruction immediately following the LW instruction gets the wrong $src1_value and $src2_value

(NOTE: This is an actual BUG and breaks functionality)

• This bug was found while checking if the above issue was causing any RAW hazards if the instruction immediately following the LW instruction accesses the destination register of the LW instruction.

• This happens because of an incorrect RF Read Bypass in the original implementation:

  $src1_value[31:0] = (>>1$rf_wr_index == $rf_rd_index1) && >>1$rf_wr_en
                      ? >>1$result : $rf_rd_data1 ;
  
  $src2_value[31:0] = (>>1$rf_wr_index == $rf_rd_index2) && >>1$rf_wr_en
                      ? >>1$result : $rf_rd_data2 ;
  

  • In this original code, the instruction immediately in the shadow of the LW instruction gets the wrong values for $src1_value, $src2_value which are the inputs to the ALU.

  • This is because, we not accounting for the fact that the data to be written to the RF could come from either the ALU ($result) or from the DMEM ($ld_data).
    $rf_wr_data[31:0] = >>2$valid_load ? >>2$ld_data : $result;

    But we are only considering the ALU output for RF Read during a RAW Hazard.

RF Read Bypass Bug

• FIX 1: During the initial debugs, I came up with the following solution to the bug based on the simulation waveforms and the VIZ_JS debug prints.

  • riscv_pipelined_withBugFix_1.tlv
    This explicitly considers the case of the instruction immediately succeeding LW.

  // Handling Read-After-Write Hazard
  $src1_value[31:0] = >>3$valid_load && (>>3$rf_wr_index == $rf_rd_index1) ? >>3$ld_data :
                      (>>1$rf_wr_index == $rf_rd_index1) && >>1$rf_wr_en ? >>1$result   :
                      $rf_rd_data1;
  
  $src2_value[31:0] = >>3$valid_load && (>>3$rf_wr_index == $rf_rd_index2) ? >>3$ld_data :
                      (>>1$rf_wr_index == $rf_rd_index2) && >>1$rf_wr_en ? >>1$result   :
                      $rf_rd_data2;
  

• FIX 2: Talking to Steve H. actually got me a better understanding of the issue, and he suggested the following code change:

  • riscv_pipelined_withBugFix_2.tlv

  // Handling Read-After-Write Hazard
  $src1_value[31:0] = (>>1$rf_wr_index == $rf_rd_index1) && >>1$rf_wr_en
                      ? >>1$rf_wr_data : $rf_rd_data1;
  
  $src2_value[31:0] = (>>1$rf_wr_index == $rf_rd_index2) && >>1$rf_wr_en
                      ? >>1$rf_wr_data : $rf_rd_data2;
  

---

Day 10: GLS of the implemented RISC-V CPU Core

• The functional verification of the design has already been completed successfully in the Makerchip IDE itself.
• To perform GLS of the implementation, we need to first convert the TL-Verilog code into synthesizable verilog and then perform the synthesis using Yosys.

10.1 Conversion of TL-Verilog code to Verilog using Sandpiper

SandPiper TL-Verilog compiler, developed by Redwood EDA can be used to convert TL-Verilog code to Verilog or SystemVerilog code. SandPiper-SaaS provides a command-line interface to run the SandPiper TL-Verilog compiler as a microservice in the cloud.

10.1.1 Sandpiper arguments

• Commonly used Sandpiper arguments:

  Argument	Details
  -p	Project name, corresponding to project configuration directory (e.g. -p verilog). (default: default)
  --hdl	The extended and target hardware description language. One of: 'verilog', 'sv'. This argument is implicit if a project (--p) argument is given. (default: sv)
  -i	Input TL-X file (with absolute or relative path). (See also, --basename.)
  -o	Produce the given translated HDL file, where the HDL language is determined from the project (-p) arg OR (--hdl ) arg. File is specified as an absolute path or a path relative to the current directory, or --outdir, if given. If a path ending in "/" or no/empty arg is given, the file name is derived from --basename or -i. A value of "STDOUT" may also be used.
  --outdir	A root directory for all produced files as a relative or absolute path.
  --inlineGen	Produce the generated code in an inline code block within the translated code, rather than in a separate file.
  --iArgs	Process command-line arguments provided in the source file.
  --bestsv	Optimize the readability/maintainability of the generated SV, unconstrained by correlation w/ TLV source.
  --noline	Disable `line directive in SV output
  --verbose	Verbose output for debug.
  --clkAlways	Use a clocking strategy for deasserted 'when' conditions (vs. clkGate and clkEnable) in which all sequential elements use an always-enabled clock with no clock gating or clock enabling. Thus, no power is saved from clock distribution and avoided logic propagation. State signal values are held using explicit recirculation.
  --clkGate	This is the default and takes precedence over the other two --clk* flags. Use a clocking strategy for deasserted 'when' conditions in which gated clock signals are provided to sequential elements. Power is saved in clock distribution and by avoiding the propagation of unused values. Explicit recirculation of state is produced only when state is consumed in the cycle that the next state value is produced.
  --clkEnable	'when' conditions are applied as clock enables to staging flip-flops. It is generally recommended to use this in combination with --clkStageAlways. This overrides --clkAlways.
  --clkStageAlways	'when' conditions such that gating/enabling is applied only to the first flip-flop after each assignment with subsequent staging using the free-running clock.

• The Sandpiper arguments could be provided either in the source file as inline arguments or in the terminal while invoking sandpiper-saas.

10.1.1.1 Providing the SandPiper arguments in the source file

• The first line of a TL-X file, called the "File Format Line", must identify the TL-X File Format Version and HDL Language, as well as a URL to the language specification.
  For example, for a TL-Verilog source file using m4 macro pre-processing language, the first line will look like as follows:
  \m4_TLV_version 1d: tl-x.org
• To this File Format line in the source file, we can add the required command-line arguments to be passed to Sandpiper, as shown in the following example:
  \m4_TLV_version 1d -p verilog --bestsv --noline --inlineGen --verbose: tl-x.org

10.1.1.2 Providing the SandPiper arguments in the command-line/ shell

• The first line in the source file - i.e., the "File Format Line" can be kept as it is and all the required arguments to sandpiper can be provided in the terminal.
  The first line will remain unchanged in the standard format as follows: \m4_TLV_version 1d: tl-x.org
• The command-line invoking sandpiper-saas now needs to include all the required arguments. So the command-line to be executed is:

  sandpiper-saas -i <input_tlv_file.tlv> -o <output_file.v> --outdir <output_dir> --bestsv --noline --inlineGen --verbose --iArgs -p verilog
  

• To specify the target HDL language, either the project (-p) argument or the target HDL (--hdl) argument can be used.
  • When the -p verilog or --hdl verilog argument is used, it needs to be provided as the last item in the command-line to avoid a bug with the argument parsing. A link to the gitlab repo with this bug fixed is provided in Method 2 below.

10.1.2 Method 1: Using the Makerchip IDE

• The Makerchip IDE provides within itself indirect access to Sandpiper - i.e., the compilation output result files can be accessed via the IDE's Editor ("E") dropdown menu.
• If the design is relatively small and not spread across multiple files, it is easier to use the Makerchip IDE itself to convert the TL-Verilog code to Verilog/ SystemVerilog.
• In this case, as we do not have direct access to sandpiper tool, all the required arguments need to be provided in the source file itself.

Steps:

1) Enter the TL-Verilog code in the Makerchip IDE's Editor interface. Modify the File Format Line in the source file to include all the sandpiper arguments that you need. 2) Under the Editor ("E") dropdown menu, choose Compile/ Sim to perform the TL-Verilog code compilation and simulation.
3.a) Under the Editor ("E") dropdown menu, choose Open Results.
3.b) Alternately, under the Editor ("E") dropdown menu, choose Show Verilog. This will open up a new webpage that shows the last compiled SystemVerilog results, along with some statistics about your TL-Verilog and SystemVerilog code.

10.1.3 Method 2: Using Sandpiper-SaaS

• Install Sandpiper-SaaS by following the steps in the following link: https://pypi.org/project/sandpiper-saas/
• Usage Examples:

  sandpiper-saas -i <input_tlv_file.tlv> -o <output_file.v> --outdir <output_dir> --bestsv --noline --iArgs --inlineGen -p verilog
  sandpiper-saas -i <input_tlv_file.tlv> -o <output_file.v> --outdir <output_dir> --bestsv --noline --iArgs --inlineGen --hdl verilog

NOTE:
There is a "bug" with argrument parsing by sandpiper-saas wherein the first argument that is not a switch (like --inlineGen, --bestsv etc.) but an optional argument (like -p verilog, --hdl verilog) gets parsed incorrectly. Because of this bug, they had to be given as the last item of the command-line as shown in the usage examples above.

The following GitLab merge request documents the issue and the fix for the same: https://gitlab.com/rweda/sandpiper-saas/-/merge_requests/2
To install sandpiper-saas with the above fix, clone the following repo https://gitlab.com/arunkumarpv/sandpiper-saas.git and install sandpiper-saas.

git clone https://gitlab.com/arunkumarpv/sandpiper-saas.git
cd sandpiper-saas
pip3 install .

10.1.4 Method 3: Using Sandpiper-SaaS with Edalize or FuseSoc

• Sandpiper-SaaS supports the Flow API and thus allowing sandpiper-saas to be used as a "frontend" to convert TL-Verilog to SystemVerilog/Verilog for any flow.
• An example of how to use sandpiper-saas with Edalize or Fusesoc in various contexts - viz. standalone tool, frontend to Vivado, in a Custom flow) is available here:
  edalize_sandpiper_example

Additional Reference Links:

1. Edalize
2. Fusesoc

10.2 GLS of the implemented RISC-V CPU Core

The TL-Verilog code of the RISC-V CPU core implementation was successfully converted to Verilog using Sandpiper-SaaS.

• To ensure that the flow is clean, I first tried GLS for a simple counter circuit:
  1. TL-V code of counter was verified in Makerchip IDE
  2. Conversion to Verilog using Sandpiper-SaaS
  3. Verify functionality using iverilog
  4. Perform synthesis using Ysosys
  5. Verify correctness by performing GLS using the synthesis output in iverilog

10.2.1 Issues

• All steps until (iv) are verified to be working fine, however, the GLS using the synthesis tool generated netlist is failing with all signals being "X".

This turned out to be due to known issues:

• OpenLane Issue #518:

  • Behavioral models in the Skywater 130 PDK are buggy and we need to switch to using the functional models by passing the FUNCTIONAL define to iverilog.
  • The UNIT_DELAY macro also needs to be set to some value.
    #iverilog -DFUNCTIONAL -DUNIT_DELAY=#1 ./lib/verilog_model/primitives.v ./lib/verilog_model/sky130_fd_sc_hd.v riscv_pipelined_Final_netlist.v tb_top_post-synth.v

• Skywater-PDK Issue #297, Skywater-PDK Issue #298:
  In sky130_fd_sc_hd.v:

  • Comment out all wire 1; statements
  • Change endif SKY130_FD_SC_HD__LPFLOW_BLEEDER_FUNCTIONAL_V to endif //SKY130_FD_SC_HD__LPFLOW_BLEEDER_FUNCTIONAL_V

With the above changes the "X"es in the simulation of the synthesized netlist is resolved and the GLS of the RISC-V CPU core is performed successfully.

iverilog -DFUNCTIONAL -DUNIT_DELAY=#1 ./lib/verilog_model/primitives.v ./lib/verilog_model/sky130_fd_sc_hd.v riscv_pipelined_Final_netlist.v tb_top_post-synth.v
./a.out
gtkwave postsynth_sim.vcd &

10.3 Results

• Yosys synthesis script

  • The synthesis script was updated at a later point in time, based on the inputs obtained from the Yosys issue raised for a problem found in post-synth STA.
    • (* keep *) attributes were added to some of the wire declarations to prevent them from getting optimized out by Yosys, abc during synthesis.
    • To avoid retaining DFF with constant inputs, the synthesis script was updated with two passes of abc.

  read_liberty -lib ./lib/sky130_fd_sc_hd__tt_025C_1v80.lib
  read_verilog ./riscv_pipelined_Final2.v
  
  hierarchy -check -top riscv_core
  synth -top riscv_core -flatten
  opt
  stat
  
  abc
  opt
  opt_clean -purge
  stat
  
  dfflibmap -liberty ./lib/sky130_fd_sc_hd__tt_025C_1v80.lib
  
  abc -liberty ./lib/sky130_fd_sc_hd__tt_025C_1v80.lib -script +strash;scorr;ifraig;retime,{D};strash;dch,-f;map,-M,1,{D}
  setundef -undriven -init -zero
  opt
  opt_clean -purge
  rename -enumerate
  stat
  
  write_verilog -noattr ./synth/riscv_pipelined_Final_netlist.v

• Generated netlist

• Statistics of synthesis output:

  === riscv_core ===
  
     Number of wires:               6845
     Number of wire bits:          10386
     Number of public wires:        6845
     Number of public wire bits:   10386
     Number of memories:               0
     Number of memory bits:            0
     Number of processes:              0
     Number of cells:               9446
       sky130_fd_sc_hd__a2111o_1       5
       sky130_fd_sc_hd__a2111oi_0     21
       sky130_fd_sc_hd__a211oi_1      11
       sky130_fd_sc_hd__a21boi_0       1
       sky130_fd_sc_hd__a21oi_1     1886
       sky130_fd_sc_hd__a221oi_1     281
       sky130_fd_sc_hd__a22o_2        30
       sky130_fd_sc_hd__a22oi_1      307
       sky130_fd_sc_hd__a31o_2         1
       sky130_fd_sc_hd__a31oi_1      259
       sky130_fd_sc_hd__a32o_1        23
       sky130_fd_sc_hd__a32oi_1        2
       sky130_fd_sc_hd__a41oi_1        1
       sky130_fd_sc_hd__and2_2         6
       sky130_fd_sc_hd__clkinv_1     847
       sky130_fd_sc_hd__dfxtp_1     1907
       sky130_fd_sc_hd__maj3_1         2
       sky130_fd_sc_hd__nand2_1     1122
       sky130_fd_sc_hd__nand3_1      115
       sky130_fd_sc_hd__nand3b_1       7
       sky130_fd_sc_hd__nand4_1      141
       sky130_fd_sc_hd__nor2_1       546
       sky130_fd_sc_hd__nor3_1        45
       sky130_fd_sc_hd__nor3b_1        1
       sky130_fd_sc_hd__nor4_1        14
       sky130_fd_sc_hd__nor4b_1        1
       sky130_fd_sc_hd__o2111a_1       3
       sky130_fd_sc_hd__o2111ai_1      6
       sky130_fd_sc_hd__o211ai_1      62
       sky130_fd_sc_hd__o21a_1         5
       sky130_fd_sc_hd__o21ai_0     1480
       sky130_fd_sc_hd__o21bai_1       7
       sky130_fd_sc_hd__o221ai_1       8
       sky130_fd_sc_hd__o22ai_1      178
       sky130_fd_sc_hd__o31ai_1        5
       sky130_fd_sc_hd__o32ai_1        1
       sky130_fd_sc_hd__o41ai_1        4
       sky130_fd_sc_hd__or2_2          5
       sky130_fd_sc_hd__or3_2          6
       sky130_fd_sc_hd__or4_2         12
       sky130_fd_sc_hd__xnor2_1       22
       sky130_fd_sc_hd__xor2_1        60
  

• Comparison of Pre-synth and Post-synth simulation results

  • The top module level input/ outputs are identical in both the Pre-Synth and Post-Synth simulation results.
  • The internal signals of relevance are also identical now that they are not optimized out due to using the (* keep *) attribute.
    

  	Comparison of Top Module Input, Outputs:
  Pre-Synth
  Post-Synth

  	Simulation time: Reset De-assertion
  Pre-Synth
  Post-Synth

  	Simulation time: Test Program Final Output
  Pre-Synth
  Post-Synth

---

---

(OLD) Results

• (OLD) Yosys synthesis script

  read_liberty -lib ./lib/sky130_fd_sc_hd__tt_025C_1v80.lib
  read_verilog ./riscv_pipelined_Final.v
  
  hierarchy -check -top riscv_core
  synth -top riscv_core
  
  dfflibmap -liberty ./lib/sky130_fd_sc_hd__tt_025C_1v80.lib
  opt
  
  abc -liberty ./lib/sky130_fd_sc_hd__tt_025C_1v80.lib -script +strash;scorr;ifraig;retime,{D};strash;dch,-f;map,-M,1,{D}
  flatten
  setundef -undriven -init -zero
  
  opt_clean -purge
  rename -enumerate
  stat
  write_verilog -noattr ./synth/riscv_pipelined_Final_netlist.v

• (OLD) Generated netlist

• (OLD) Statistics of synthesis output:

  === riscv_core ===
  
     Number of wires:               7014
     Number of wire bits:          10175
     Number of public wires:        7014
     Number of public wire bits:   10175
     Number of memories:               0
     Number of memory bits:            0
     Number of processes:              0
     Number of cells:               9049
       sky130_fd_sc_hd__a2111oi_0     20
       sky130_fd_sc_hd__a211oi_1      17
       sky130_fd_sc_hd__a21boi_0       9
       sky130_fd_sc_hd__a21o_2         1
       sky130_fd_sc_hd__a21oi_1      917
       sky130_fd_sc_hd__a221oi_1     238
       sky130_fd_sc_hd__a22o_2        27
       sky130_fd_sc_hd__a22oi_1      479
       sky130_fd_sc_hd__a311oi_1       5
       sky130_fd_sc_hd__a31oi_1      738
       sky130_fd_sc_hd__a32o_1         3
       sky130_fd_sc_hd__a32oi_1        3
       sky130_fd_sc_hd__and2_2        12
       sky130_fd_sc_hd__and3_2         2
       sky130_fd_sc_hd__clkinv_1     784
       sky130_fd_sc_hd__dfxtp_1     1962
       sky130_fd_sc_hd__maj3_1         2
       sky130_fd_sc_hd__mux2i_1        3
       sky130_fd_sc_hd__nand2_1      860
       sky130_fd_sc_hd__nand3_1      247
       sky130_fd_sc_hd__nand3b_1       3
       sky130_fd_sc_hd__nand4_1      175
       sky130_fd_sc_hd__nor2_1       573
       sky130_fd_sc_hd__nor3_1        52
       sky130_fd_sc_hd__nor4_1        46
       sky130_fd_sc_hd__nor4b_1        2
       sky130_fd_sc_hd__o2111ai_1     17
       sky130_fd_sc_hd__o211a_1        1
       sky130_fd_sc_hd__o211ai_1      46
       sky130_fd_sc_hd__o21a_1         3
       sky130_fd_sc_hd__o21ai_0     1476
       sky130_fd_sc_hd__o21bai_1      29
       sky130_fd_sc_hd__o221ai_1       5
       sky130_fd_sc_hd__o22ai_1      183
       sky130_fd_sc_hd__o2bb2ai_1      2
       sky130_fd_sc_hd__o311a_2        1
       sky130_fd_sc_hd__o311ai_0       9
       sky130_fd_sc_hd__o31ai_1        1
       sky130_fd_sc_hd__o41ai_1        1
       sky130_fd_sc_hd__or2_2         19
       sky130_fd_sc_hd__or2b_2         1
       sky130_fd_sc_hd__or3_2          1
       sky130_fd_sc_hd__or4_2          1
       sky130_fd_sc_hd__xnor2_1       25
       sky130_fd_sc_hd__xor2_1        48
  

• (OLD) Comparison of Pre-synth and Post-synth simulation results

  • The top module level input/ outputs are identical in both the Pre-Synth and Post-Synth simulation results.
  • However, on a closer look, due to the fact that we are hardcoding the test program in the RTL, the synthesis tools optimizes out some of the constant signals that are unused. Hence, the values of all of the interal signals may not necessarily match between Pre-Synth and Post-Synth simulation.
    
    For example, the signal CPU_imem_rd_data_a1 is different in the two due to the fact that the instructions to be executed are hardcoded as constants and not all of the bits in the 32-bit Instruction are necessary for decoding.
    
    

  	(OLD) Comparison of Top Module Input, Outputs:
  Pre-Synth
  Post-Synth

  	(OLD) Comparison of some Internal Signals:
  Pre-Synth
  Post-Synth

---

Days 11-13: Advanced Synthesis and STA

Day 11: Introduction to STA

Static Timing Analysis (STA) is a method of validating the timing performance of a design by checking all possible paths for timing violations.

• The design is broken down into timing paths having start and endpoints, the signal propagation delay along each path is calculated, and checked for violations of timing constraints inside the design and at the input/output interfaces.
• The STA analysis is the static type - i.e., the timing analysis is carried out statically and does not depend upon the data values being applied at the input pins.

As the naming suggests, there is another method - Dynamic Timing Analysis (DTA) where the timing behaviour is analyzed by applying a stimulus/ test vector at the input signals, the resulting behaviour of the design is captured in the simulation and functionality is verified using the testbenches.

• This approach verifies the design meets the timing requirements while simultaneously ensuring the functionality.
• However, this is a very expensive method in terms of both simuation run time and resources as we cannot verify the entire space of test vectors exhaustively.
• A trade-off could be to select the test vectors that exercise majority of the circuits, but this is usually not sufficient for sign-off as the coverage/ confidence obtained is only as exhaustive as the test vectors used and no 100% guarantee that the design will work across conditions.

STA Basics

STA can performed at various stages of the ASIC design flow, with the emphasis being on different aspects.

Timing Arcs

1. Combinational arcs: Between input and output pin of a combinational block/ cell.
2. Sequential arcs: Between the clock pin and either the input or output
   • Timing check arc: Between the clock pin and the input. (For example, the setup and hold timing arcs between the clock pin and input data pin of a Flip Flop)
   • Delay arc: Between the clock pin and the output.
3. Net arcs: Between the output pin of a cell to the input pin of another cell. (i.e., between the driver pin of a net and the load pin of that net)

Unateness

Defines how the output changes for different types of transitions on the input.

1. Positive unate: Rising input transition causes rising output transition and falling input transition causes falling output transition.
   • Examples: Buffer, AND Gate, OR Gate
2. Negative unate: Rising input transition causes falling output transition and falling input transition causes rising output transition.
   • Examples: Inverter, NAND Gate, NOR Gate
3. Non-unate: The output transition is determined not only by the direction of an input but also on the state of the other inputs.
   • Examples: XOR Gate

Advanced Synthesis and STA with DC

To Do: Complete the documentation of the theoretical aspects

STA of the RISC-V CPU core using OpenSTA

• Design Constraints

  set_units -time ns
  
  set period 10.000
  create_clock -name clk -period $period [get_ports {clk}]
  
  set_clock_latency -source -min 1 clk
  set_clock_latency -source -max 4 clk
  
  set clk_uncertainty_factor_setup 0.05
  set clk_uncertainty_setup [expr $period * $clk_uncertainty_factor_setup]
  set clk_uncertainty_factor_hold 0.02
  set clk_uncertainty_hold [expr $period * $clk_uncertainty_factor_hold]
  set_clock_uncertainty -setup $clk_uncertainty_setup [get_clock clk]
  set_clock_uncertainty -hold $clk_uncertainty_hold [get_clock clk]
  
  set min_input_dly_factor 0.1
  set max_input_dly_factor 0.3
  set min_input_dly [expr $period * $min_input_dly_factor]
  set max_input_dly [expr $period * $max_input_dly_factor]
  set_input_delay -clock clk -min $min_input_dly [get_ports reset]
  set_input_delay -clock clk -max $max_input_dly [get_ports reset]
  
  
  set min_tran_factor 0.01
  set max_tran_factor 0.05
  set min_tran [expr $period * $min_tran_factor]
  set max_tran [expr $period * $max_tran_factor]
  set_input_transition -max $min_tran [get_ports reset]
  set_input_transition -min $max_tran [get_ports reset] 
  
  set min_ouput_dly_factor 0.2
  set max_ouput_dly_factor 0.5
  set min_ouput_dly [expr $period * $min_ouput_dly_factor]
  set max_ouput_dly [expr $period * $max_ouput_dly_factor]
  set_output_delay -clock clk -min $min_ouput_dly [get_ports out]
  set_output_delay -clock clk -min $max_ouput_dly [get_ports out]
  

• SDC File written from OpenSTA using write_sdc command

• OpenSTA shell showing snapshots The following snapshots show the commands to:

  • read_liberty: Read the Liberty format library file
  • read_verilog: Read the gate level verilog netlist file
  • link_design: Link (elaborate, flatten) the the top level cell
  • read_sdc: Read SDC commands from the given constraints file
  • check_setup: Perform sanity checks on the design
  • report_checks: Report paths in the desing based on the additional arguments/ filters provided

OpenSTA commands to do STA

(OLD) Min path	(OLD) Max path

• As can be seen from the OpenSTA log, the design has 3 unconstrained endpoints.

  _15766_/D
  _15768_/D
  _15777_/D
  

  • On inspecting the gate level netlist, these endpoints are the D input pins of DFF (sky130_fd_sc_hd__dfxtp_1) tied to a constant "0".
  • Yosys should have reduced them to constant drivers but somehow they still remain.
  • A Yosys GitHub issue has been raised to check if there is a way to avoid this.
    Details of the issue can be read here: YosysHQ/yosys#4266

• The design also fails to meet the target timing requirements:

  • Min path: most probably due to the overly conservative/ pessimistic values of the clock source latencies used.
  • Max path: due to the conservative values of clock source latencies used and undefined fanout constraints. (Inverter instance _10539_ of type sky130_fd_sc_hd__clkinv_1 is having a fanout of 932)

UPDATE:

• The synthesis script was updated based on inputs from Yosys issue #4266.
  • The DFF with constant inputs have been taken care of.
  • The STA analysis of the new netlist is given below:

Min path	Max path

---

Day 12: Advanced SDC Constraints

---

Day 13: STA using OpenSTA

---

Days 14-18: Circuit Design using SKY130 PDK

Day 14: CMOS Fundamentals

14.1 What is SPICE and why do we need SPICE simulations ?

14.1.1 What is SPICE ?

Fabricating ICs is very expensive and time-consuming, so designers need simulation tools to explore the design space and verify designs before they are fabricated. Simulation is cheap, but silicon revisions (even a single Metal layer change) are prohibitively expensive.

Simulators operate at many levels of abstraction, from process through architecture.

• Process simulators such as SUPREME predict how factors in the process recipe such as time and temperature affect device physical and electrical characteristics.
• Circuit simulators such as SPICE and Spectre use device models and a circuit netlist to predict circuit voltages and currents, which indicate performance and power consumption.
• Logic simulators such as VCS and ModelSim are widely used to verify correct logical operation of designs specified in a hardware description language (HDL).
• Architecture simulators, sometimes offered with a processor’s development toolkit, work at the level of instructions and registers to predict throughput and memory access patterns, which influence design decisions such as pipelining and cache memory organization.

The various levels of abstraction offer trade-offs between degree of detail and the size of the system that can be simulated. VLSI designers are primarily concerned with circuit and logic simulation.

SPICE (Simulation Program with Integrated Circuit Emphasis) was originally developed in the 1970s at University of California, Berkeley. It solves the nonlinear differential equations describing components such as transistors, resistors, capacitors, and voltage sources.

Based on the original SPICE, there are many SPICE versions available - both free (like Ngspice, Xyce, LTSpice, TINA-TI) as well as commercial (HSPICE, PSPICE). All versions of SPICE read an input file and generate an output with results, warnings, and error messages. The input file is often called a SPICE deck and each line is called a card because it was once provided to a mainframe as a deck of punch cards.

A circuit simulator is provided with an input file that contains:

• A netlist consisting of components and nodes detailing the circuit connectivity.
  The netlist can be entered by hand or extracted from a circuit schematic or layout in a CAD program.
• Component behaviour by means of device models and model parameters.
• The Initial state of the circuit -- initial conditions
• Inputs to the circuit, called stimulus
• Simulation options & analysis commands that explain the type of simulation to be run.

Ref: CMOS VLSI Design - A Circuits and Systems Perspective - Weste & Harris

Analysis Types supported by SPICE:

Analysis Type	Details
DC Analysis	Find the DC operating point of the circuit i.e., all voltages and currents
AC Small-Signal Analysis	AC analysis is limited to analog nodes and represents the small signal, sinusoidal solution of the analog system described at a particular frequency or set of frequencies.
Transient Analysis	Transient analysis is an extension of DC analysis to the time domain. In other words, it solves a DC Analysis for each timestep based on initial conditions.
Pole-Zero Analysis	Computes the poles and/or zeros in the small-signal ac transfer function.
Small-Signal Distortion Analysis	Computes steady-state harmonic and intermodulation products for small input signal magnitudes.
Sensitivity Analysis	Calculate either the DC operating-point sensitivity or the AC small-signal sensitivity of an output variable with respect to all circuit variables, including model parameters.
Noise Analysis	Measures the device-generated noise for a given circuit.

The following images show how a SPICE deck is written to perform DC analysis of an NMOS transistor:

14.1.2 Why SPICE in VLSI Design ?

In Digital VLSI Design, the timing, power, process variation, noise and signal integrity analyses all ultimately rely on SPICE for accurate modelling and characterization of the various Standard Cells and macros used in the design.

14.2 NMOS Transistor

14.2.1 (Planar) NMOS Transistor Structure

14.2.2 Transistor Operation: Cut-off Region, Surface Inversion & Threshold Voltage

• Without the application of a Gate potential, the transistor is said to be in Cut-off region as there is no conducting path between the Source and Drain terminals.
• On the application a sufficiently high Gate-to-Source voltage, a conductive channel starts to form underneath the Gate composed of minority carriers (electrons in an NMOS) and at a certain voltage called the Threshold voltage, surface inversion occurs when the concentration of minority carriers in the channel becomes equal to the concentration of majority carriers in the bulk.
• When VGS is increased further by several $\Phi_t (=thermal voltage, \dfrac{k_B T}{q}) $, the transistor moves into the strong inversion region. Here, the minority carrier concentration in the channel is a strong function of the applied gate potential.

The below images depict the same for an NMOS transistor:

14.2.3 Effect of Subtrate/ Body Bias on Threshold Voltage

• The Source-to-Substrate amd Drain-to-Substrate pn junctions must always be reverse biased for the "normal operation" of the MOS transistor, so $V_{SB}, V_{DB}$ must always be greater than or equal to zero for an NMOS transistor.
• If $V_{SB} = 0$, then surface inversion is achieved at a Gate-to-Source voltage equal to VT0
• However, when $V_{SB} &gt; 0$, the electrons from the channel can move laterally and flow out of the source terminal resulting in a reduced carrier concentration in the channel.
• Thus, with when $V_{SB} &gt; 0$, a higher Gate-to-Source voltage is required to achieve surface inversion.
• In other words, the threshold voltage of an NMOS increases when $V_{SB} &gt; 0$.

Threshold Voltage Equation considering Body Bias:

14.2.4 Resistive/ Linear/ Triode Region of Operation

Let us analyse the condition when we apply a Gate-Source potential, $V_{GS} &gt;= V_{TH}$ and a small value of $V_{DS}$ is applied across the channel from Drain-to-Source.

14.2.4.1 Derivation of Drain Current Equation

Using a simple first-order analysis, let us try to derive an equation for the Drain Current, $I_D$ that results due to the $V_{GS}$ and $V_{DS}$ values applied.

The mechanism for the Drain current, $I_D$ is carrier drift under the lateral electric field due to $V_{DS}$.

• Let $V_{GS}$ be held constant at a value greater than $V_{TH}$.

• The applied $V_{DS}$ appears as a voltage drop across the length of the channel.

• As we travel along the channel from Source to Drain, the voltage (measured relative to the Source terminal) increases from zero to $V_{DS}$.

• Thus the voltage between the gate and points along the channel decreases from $V_{GS}$ at the Source end to $V_{GD} = V_{GS}-V_{DS}$ at the Drain end.

• At a point x along the channel, the voltage is $V(x)$, and the gate-to-channel voltage at that point equals $V_{GS} – V(x)$.

Under the assumption that this voltage exceeds the threshold voltage all along the channel, the induced channel charge per unit area at point x can be computed.

$Q_i(x) = -C_{ox} [V_{GS} - V(x) -V_{TH}]$
$where:$
$~~~~~~~~ C_{ox} = \dfrac{\epsilon_{ox}}{t_{ox}}$

The current is given as the product of the drift velocity of the carriers, $v_n$ and the available charge. Due to charge conservation, it is a constant over the length of the channel. W is the width of the channel in a direction perpendicular to the current flow.

$I_D = -v_n(x) * Q_i(x) * W$

The electron velocity is related to the electric field through a parameter called the mobility $\mu_n$ (expressed in $\dfrac{m^2}{V.s}$).

Drift velocity, $v_n = -\mu_n \dfrac{dV}{dx}$

$\therefore I_D = -\mu_n \dfrac{dV}{dx} * -C_{ox} [V_{GS} - V(x) - V_{TH}] * W$

$i.e., I_D dx = \mu_n C_{ox} W [V_{GS} - V(x) -V_{TH}] dV$

Integrating the equation over the length of the channel L yields the voltage-current relation of the transistor:
$\boxed{I_D = {k_n}^\prime \dfrac{W}{L} \left[ (V_{GS}-V_{TH})V_{DS} - \dfrac{{V_{DS}}^2}{2} \right] = k_n \left[ (V_{GS}-V_{TH})V_{DS} - \dfrac{{V_{DS}}^2}{2} \right]}$

$where:$
$~~~~~~~~ {k_n}^\prime$ is the process transconductance parameter.
$~~~~~~~~ {k_n}^\prime = \mu_nC_{ox}$

The product of process transconductance, ${k_n}^\prime$ and the $\dfrac{W}{L}$ ratio of the transistor is called the gain factor, $k_n$ of the device.

$~~~~~~~~ k_n = {k_n}^\prime * \left( \dfrac{W}{L} \right)$

Now, the above equation for Drain Current:
$I_D = k_n * \left[(V_{GS} - V_{TH}) * V_{DS} - \dfrac{V_{DS}^2}{2}\right]$ is a quadratic function of $V_{DS}$.

But at low values, the $\dfrac{V_{DS}^2}{2}$ term can be ignored as it is close to zero. Hence the equation for $I_D$ can be approximated to be a linear function of $V_{DS}$:

$\boxed {i.e., I_D = k_n * (V_{GS} - V_{TH}) * V_{DS}} ~~~~ $ Hence the name Linear Region

For the example scenario we were discussing, this translates to:

14.2.5 Saturation/ Pinch-Off Region of Operation

• When a $V_{DS}$ voltage is applied, the channel voltage becomes a function of both $V_{GS}$ and $V_{DS}$.
• Since the induced channel depth depends on the channel voltage relative to the Gate terminal, and specifically on the amount by which this voltage exceeds the threshold voltage, $V_{TH}$, we find that the channel is no longer of uniform depth; rather, the channel will take a tapered shape:
  • being deepest at the Source end, where the depth is proportional to $[V_{GS}-V_{TH}]$, and
  • shallowest at the drain end, where the depth is proportional to $[V_{GS}-V_{TH}-V_{DS}]$.

14.2.5.1 Pinch-off Region Condition

• As the value of the Drain-Source voltage is increased further, the assumption that the channel voltage is larger than the threshold all along the channel ceases to hold.
  • In the limiting case, the channel depth at the drain end reduces to zero and the channel is said to be "pinched-off". This happens when $V_{GD}$ is just equal to the threshold voltage, $V_{TH}$.
  • i.e., $V_{DS}= V_{GS}-V_{TH} ~~~~~~~~ (=V_{OV})$, Gate Over-drive voltage
• At this point, the induced charge is zero, and the conducting channel disappears or is pinched off starting from the Drain end.

14.2.5.2 Drain current model for Saturation Region of Operation

• Increasing $V_{DS}$ beyond the $V_{DSsat}$ value of ($V_{GS}-V_{TH}$) has no effect on the channel shape and charge.

• Thus, the current through the channel remains constant at the value reached for $V_{DS}= V_{GS}-V_{TH}$.

• The MOSFET is said to have entered saturation/ pinch-off regsion at:

  • $V_{DS} = V_{DSsat} = V_{GS}-V_{TH}$

• Subsequently, the Saturation Drain current:

  • $I_D = I_{Dsat} = {k_n}^\prime \dfrac{W}{L} \left[ (V_{GS}-V_{TH})V_{DSsat} - \dfrac{{V_{DSsat}}^2}{2} \right]$
  • $\boxed{i.e., I_{Dsat} = \dfrac{1}{2} {k_n}^\prime \dfrac{W}{L} {\left[ {V_{GS}-V_{TH}}^2\right]}}$

• Channel pinch-off does not imply channel blockage

  • Current continues to flow through the pinched-off channel.
  • The electrons that reach the drain end of the channel are accelerated through the depletion region that exists there and into the drain terminal.
  • Any increase in $V_{DS}$ above $V_{DSsat}$ appears as a voltage drop across the depletion region
  • Thus, both the current through the channel, $I_{Dsat}$ and the voltage drop across the channel $i.e., V_{DSsat} = (V_{GS}-V_{TH})$ remain constant in saturation.

14.2.5.3 Channel Length Modulation and Finite Output Resistance in Saturation

• The above Drain current equation seems to indicate that in sarutation, $I_D$ is independent of $V_{DS}$

• Thus a change, $\Delta V_{DS}$, in the Drain-to-Source voltage causes a zero change in $I_D$, which implies that the incremental resistance looking into the drain of a saturated MOSFET is infinite.

• This, however, is an idealization based on the premise that once the channel is pinched off at the drain end, further increases in $V_{DS}$ have no effect on the channel’s shape.

• But, in practice, increasing $V_{DS}$ beyond $V_{GS}-V_{TH}$ does affect the channel somewhat.

  • As $V_{DS}$ is increased, the channel pinch-off point is moved slightly away from the drain, toward the source.
  • That is, the voltage across the channel remains constant at $V_{GS}-V_{TH}$, and
  • The additional voltage applied to the drain appears as a voltage drop across the narrow depletion region between the end of the channel and the drain region.
  • This voltage accelerates the electrons that reach the drain end of the channel and sweeps them across the depletion region into the drain.

• Note, however, that (with depletion-layer widening) the channel length is reduced from $L$ to $(L-\Delta L)$, called Channel Length Modulation (CLM).

• Since $I_D$ is inversely proportional to the channel length, $I_D$ increases with $V_{DS}$.

• CLM can be accounted for in the expression for $I_D$ by including a factor $[1 + \lambda (V_{DS}-V_{DSsat})]$.
  For simplicity, we use: $[1 + \lambda V_{DS}]$

  $\boxed{\therefore I_{D} = \dfrac{1}{2} {k_n}^\prime \dfrac{W}{L} \left[ {(V_{GS}-V_{TH}}^2\right] (1 + \lambda V_{DS})}$

• The CLM parameter, $\lambda$ is a device parameter having the units of $V^{-1}$ . Its value depends both on the process technology used to fabricate the device and on the channel length, $L$ that the circuit designer selects.

• Output Resistance, $r_o = \dfrac{1}{\lambda I_D}$

14.2.6 Lab: ID vs. VDS for different VGS - sky130 NMOS (W=5um, L=2um)

SPICE File: day1_nfet_idvds_L2_W5.spice

*** Model Description ***
.param temp=27

*** Including sky130 library files ***
.lib "sky130_fd_pr/models/sky130.lib.spice" tt

*** Netlist Description ***
XM1 vdd n1 0 0 sky130_fd_pr__nfet_01v8 w=5 l=2
R1 n1 in 55
Vdd vdd 0 1.8
Vin in 0 1.8

*** Simulation Commands ***
.op
.dc Vdd 0 1.8 0.1 Vin 0 1.8 0.2

.control
run
display
setplot dc1
.endc

.end

Output:

---

Day 15: Velocity Saturation and CMOS Inverter VTC

15.1 Velocity Saturation

• The behavior of transistors with very short channel lengths (called short-channel devices) deviates considerably from the resistive and saturated models.
• The main reason for this deviation is the velocity saturation effect.
• We had seen previously that the drift velocity is modelled by:
  Drift velocity, $v = -\mu \dfrac{dV}{dx}$
  • i.e., the velocity of the carriers is proportional to the electrical field, independent of the value of that field. In other words, the carrier mobility is a constant.
• However, at high electric field strengths, the carriers fail to follow this linear model.
• When the electrical field along the channel reaches a critical value $E_c$, the velocity of the carriers tends to saturate due to scattering effects (collisions suffered by the carriers).

• For p-type silicon:
  • the critical field at which electron saturation occurs is around $1.5 \times 10^6 V/m ~ (or~1.5 V/\mu m)$, and
  • the saturation velocity $v_{sat}$ approximately equals $10^5 m/s$
• This means that in an NMOS device with a channel length of $1~\mu m$, only a couple of volts of $V_{DS}$ is needed to reach the electron velocity saturation point. This condition is easily met in current short-channel devices.
• Holes in a n-type silicon saturate at the same velocity, although a higher electrical field is needed to achieve saturation. Velocity-saturation effects are hence less pronounced in PMOS transistors.
• The drift velocity can be roughly approximated as a piece-wise linear function of the electrical field as follows:
  • $v = \dfrac{\mu E}{1+(E/E_c)}~~~~~,for~ E \le E_c$
    $~~~=v_{sat}~~~~~~~~~~~~~~~~~~~~,for~E &gt; E_c$
• For continuity at $E=E_c$, we get: $E_c = \dfrac{2v_{sat}}{\mu}$

15.1.1 Drain Current in Resistive/ Linear Region

Now the drain current equation in the resistive region can be re-evaluated using:

$I_D = -v_n(x) * Q_i(x) * W$
$I_D = -v_n(x) * -C_{ox} [V_{GS} - V(x) -V_{TH}] * W$
$v_n = \dfrac{\mu_n E}{1+(E/E_c)}$

Now, $E = \dfrac{dV}{dx}$

$\therefore v_n = \dfrac{\mu_n (dV/dx)}{1+(1/E_c)(dV/dx)}$

$\therefore I_D = \dfrac{\mu_n (dV/dx)}{1+(1/E_c)(dV/dx)} * C_{ox} [V_{GS} - V(x) - V_{TH}] * W$

$i.e., I_D \left[ 1+\dfrac{1}{E_c}\dfrac{dV}{dx}\right] = \mu_n \dfrac{dV}{dx} * C_{ox} [V_{GS} - V(x) - V_{TH}] * W$

Integrating w.r.t $x$ from $x=0 ~~ (where ~~ V=0)$ to $x=L ~~ (where ~~ V=V_{DS})$, we get:

$I_D \left[ L+(V_{DS}/E_c)\right] = \mu_n C_{ox} [(V_{GS} - V_{TH})V_{DS} - \dfrac{V_{DS}^2}{2} ] * W$

Re-arranging:
$I_D = \left[\dfrac{1}{L+(V_{DS}/E_c)}\right] \mu_n * C_{ox} [(V_{GS} - V_{TH})V_{DS} - \dfrac{V_{DS}^2}{2} ] * W$

$i.e., I_D = \left[\dfrac{1}{1+(V_{DS}/E_c L)}\right] \mu_n * C_{ox} [(V_{GS} - V_{TH})V_{DS} - \dfrac{V_{DS}^2}{2} ] * \dfrac{W}{L}$

$i.e., \boxed{I_D = \left[\dfrac{1}{1+(V_{DS}/E_c L)}\right]~\mu_n C_{ox} \dfrac{W}{L} [(V_{GS} - V_{TH})V_{DS} - \dfrac{V_{DS}^2}{2} ]}$

$~~~~~~~~~~~~\boxed{I_D = \kappa(V_{DS}) \mu_n C_{ox} \dfrac{W}{L}\left[(V_{GS}-V_{TH})V_{DS} - \dfrac{{V_{DS}^2}}{2} \right]}$

$~~~~~~~~~~~~where, ~~~~ \kappa(V_{DS}) = \dfrac{1}{1+(V_{DS}/E_c L)}$

• $\kappa$ is a measure of the velocity saturation since $V_{DS}/L$ is the average field in the channel.
• In the case of long-channel devices (where $L$ is large), or when the value of $V_{DS}$ is small, $\kappa$ approaches 1 and the above equation simplifies to the traditional equation we had derived first using the constant mobility model.
• For short-channel devices, $\kappa$ is smaller than 1, implying the delivered current is smaller than what would be normally expected.

15.1.2 Drain Current in Saturation Region

Coming to the Drain current in saturation region:

$I_D = -v_n(x) * -C_{ox} [V_{GS} - V(x) -V_{TH}] * W ~~~~~~~~ |with ~~ v_n(x)=v_{sat} ~~ and ~~ V(x)=V_{DSAT}$
$i.e., \boxed{I_{DSAT} = v_{sat} C_{ox} W [(V_{GS} - V_{TH}) - V_{DSAT}]}$

$I_{DSAT}$ can also be evaluated by replacing $V_{DS}=V_{DSAT}$ in the linear region equation derived in the previous section.

$\therefore ~~~~ \boxed{I_{DSAT} = \kappa(V_{DSAT}) \mu_n C_{ox} \dfrac{W}{L}\left[(V_{GS}-V_{TH})V_{DSAT} - \dfrac{{V_{DSAT}^2}}{2} \right]}$

Equating these two expressions for $I_{DSAT}$ to solve for $V_{DSAT}$, we get:
$I_{DSAT} = v_{sat} C_{ox} W [(V_{GS} - V_{TH}) - V_{DSAT}]~= \kappa(V_{DSAT}) \mu_n C_{ox} \dfrac{W}{L}\left[(V_{GS}-V_{TH})V_{DSAT} - \dfrac{{V_{DSAT}^2}}{2} \right]$

$i.e.,$
$\dfrac{\mu_n E_c}{2} C_{ox} W [(V_{GS} - V_{TH}) - V_{DSAT}] = \kappa(V_{DSAT}) \mu_n C_{ox} \dfrac{W}{L}\left[(V_{GS}-V_{TH})V_{DSAT} - \dfrac{{V_{DSAT}^2}}{2} \right]$

$E_c L [(V_{GS} - V_{TH}) - V_{DSAT}] = \kappa(V_{DSAT}) \left[2(V_{GS}-V_{TH})V_{DSAT} - {V_{DSAT}^2} \right]$

This can be further simplified to the below:
$~~~~~~~~ \boxed{V_{DSAT} = \kappa(V_{GT}) ~ V_{GT}} ~~~~ , where ~ V_{GT} = (V_{GS} - V_{TH})$

This has the following implications:

1. The saturation current $I_{DSAT}$ displays a linear dependence with respect to the Gate-to-Source voltage $V_{GS}$, which is in contrast with the squared dependence in the long-channel devices. This reduces the amount of current a transistor can deliver for a given control voltage.

2. For a short-channel device and for large enough values of $V_{GT}$, $\kappa(V_{GT})$ is substantially smaller than 1, hence $V_{DSAT} &lt; V_{GT}$. The device enters saturation before $V_{DS}$ reaches $(V_{GS} - V_{TH})$. Short-channel devices therefore experience an extended saturation region and tend to operate more often in saturation region compared to long-channel devices.

15.1.3 Unified MOS Model for Manual Analysis

• It is desrirable to have a mathematical model that abstracts the behavior of the MOS transistor into a simple and tangible analytical model that does not lead to hopelessly complex equations, yet captures the essentials of the device.
• This is required so that the designer is able to have an intuitive insight into the behaviour of a circuit and how the design parameters affect its operation.

• The first-order expressions derived earlier can be combined into a single unified model that presents the transistor as a single current source the value as defined below:
  • The model employs the voltages at the four terminals of the transistor, along with a set of five parameters: $V_{TO}, ~ \gamma, ~ V_{DSAT}, ~ k^{\prime}, ~ and ~ \lambda$.

The following slides show how the unified model works in the different Regions of Operation of the MOS transistor:

15.1.4 Lab: Velocity Saturation - ID vs. VDS - tsmc 0.25um

SPICE File: nmos_chara_W1.8u_L1.2u.spice

*** Netlist Description ***
M1 vdd n1 0 0 nmos W=1.8u L=1.2u
R1 in n1 55
Vdd vdd 0 2.5
Vin in 0 2.5

*** .include model ***
.lib "tsmc_025um_model.mod" cmos_models

*** Simulation Commands ***
.op
.dc Vdd 0 2.5 0.1 Vin 0 2.5 0.5

.control
run
display
setplot dc1
plot -vdd#branch
.endc

.end

SPICE File: nmos_chara_W0.375u_L0.25u.spice

*** Netlist Description ***
M1 vdd n1 0 0 nmos W=0.375u L=0.25u
R1 in n1 55
Vdd vdd 0 2.5
Vin in 0 2.5

*** .include model ***
.lib "tsmc_025um_model.mod" cmos_models

*** Simulation Commands ***
.op
.dc Vdd 0 2.5 0.1 Vin 0 2.5 0.5

.control
run
display
setplot dc1
plot -vdd#branch
.endc

.end

Output:

15.1.5 Lab: Velocity Saturation - sky130 (W=0.39um, L=0.15um)

SPICE File: day2_nfet_idvds_L015_W039.spice

*** Model Description ***
.param temp=27

*** Including sky130 library files ***
.lib "sky130_fd_pr/models/sky130.lib.spice" tt

*** Netlist Description ***

XM1 Vdd n1 0 0 sky130_fd_pr__nfet_01v8 w=0.39 l=0.15
R1 n1 in 55
Vdd vdd 0 1.8V
Vin in 0 1.8V

*** Simulation commands ***
.op
.dc Vdd 0 1.8 0.1 Vin 0 1.8 0.2

.control
run
display
setplot dc1
.endc

.end

Output:

15.1.6 Lab: Velocity Saturation - ID$ vs. VGS - sky130 NMOS (5u/2u vs. 0.39u/0.15u)

SPICE File: day2_nfet_idvgs_L015_W039.spice

*** Model Description ***
.param temp=27

*** Including sky130 library files ***
.lib "sky130_fd_pr/models/sky130.lib.spice" tt

**** Netlist Description ***
XM1 Vdd n1 0 0 sky130_fd_pr__nfet_01v8 w=0.39 l=0.15
R1 n1 in 55
Vdd vdd 0 1.8V
Vin in 0 1.8V

*** Simulation Commands ***
.op
.dc Vin 0 1.8 0.1 Vdd 1.8 1.8 0.2 

.control
run
display
setplot dc1
.endc

.end

Output:

15.2 CMOS Inverter VTC

• The Voltage-Transfer Characteristic (VTC) can be graphically derived by superimposing the current characteristics of the NMOS and the PMOS devices -- i.e., their respective load-line plots.

• It requires that the I-V curves of the NMOS and PMOS devices are transformed onto a common coordinate set.

• To plot the VTC or the Input-Output transfer characteristics ($V_{out}$ vs. $V_{in}$) of the CMOS Inverter, let us choose the input voltage $V_{in}, the output voltage $V_{out}$ and the NMOS drain current $I_{DSn}$ as the independent variables.

• The PMOS I-V relations can be translated into this variable space by the following relations:

KCL, KVL Constraints
$I_{DSp} = -I_{DSn}$
$V_{GSn} = V_{in}; ~~~~~~~~ V_{GSp} = V_{in}-V_{DD}$
$V_{DSn} = V_{out}; ~~~~~~~ V_{DSp} = V_{out}-V_{DD}$

• Now the input voltage, $V_{in}$ needs to be swept from 0 to $V_{DD}$ and find out the corresponding values of $V_{out}$.
• The following transformations adjust the original PMOS I-V curves to the chosen common coordinate set {$V_{in}$, $V_{out}$ and $I_{Dsn}$}.

Transform	Comments
$I_{DSp} ~ \longrightarrow ~ I_{DSn}$	Reflection about x-axis of $I_{DSp} ~ vs. ~ V_{DSp}$ curve. where, $I_{DSn} = -I_{DSp}$
$V_{GSp} ~ \longrightarrow ~ V_{in}$	Variable change from $V_{GSp}$ to $V_{in}$. where, $V_{in} = V_{GSp}+V_{DD}$
$V_{DSp} ~ \longrightarrow ~ V_{out}$	Translation along the x-axis. $V_{out} = V_{DSp}+V_{DD}$

• Now, the PMOS and NMOS load lines are overlaid on top of each other.
• For a DC operating points to be valid, the currents through the NMOS and PMOS devices must be equal. Graphically, this means that the DC operating points must be located at the intersection of corresponding load lines.
• Find the set of all $(V_{in}, V_{out})$ pairs corresponding to the points of intersection and plot them to generate the VTC of the CMOS Inverter circuit.

Relationships between voltages for the three regions of operation of PMOS, NMOS in a CMOS inverter:

	Cutoff	Linear	Saturation
NMOS	$V_{GSn} < V_{Tn}$ $V_{in} < V_{Tn}$	$V_{GSn} > V_{Tn}$ $V_{in} > V_{Tn}$ $V_{DSn} < (V_{GSn}-V_{Tn})$ $V_{out} < (V_{in}-V_{Tn})$	$V_{GSn} > V_{Tn}$ $V_{in} > V_{Tn}$ $V_{DSn} > (V_{GSn}-V_{Tn})$ $V_{out} > (V_{in}-V_{Tn})$
PMOS	$V_{GSp} > V_{Tp}$ $V_{in} > V_{DD} - \mid V_{Tp} \mid$	$V_{GSp} < V_{Tp}$ $V_{in} < V_{DD}-\mid V_{Tp} \mid$ $V_{DSp} > (V_{GSp}-V_{Tp})$ $V_{out} > (V_{in}+\mid V_{Tp} \mid)$	$V_{GSp} < V_{Tp}$ $V_{in} < V_{DD}-\mid V_{Tp} \mid$ $V_{DSp} < (V_{GSp}-V_{Tp})$ $V_{out} < (V_{in}+\mid V_{Tp} \mid)$

Note: $V_{Tp}$ is negative.

• The operation of the CMOS inverter can be divided into five regions - A, B, C, D, E - indicated in the above figure.
• For simplicity, let us assume $V_{Tp} = –V_{Tn}$ and that the PMOS transistor is ~2–3 times as wide as the NMOS transistor so that $k_n = k_p$.

Region	Condition	PMOS	NMOS	Output
A	$0 \le V_{in} < V_{Tn}$	Linear	Cutoff	$V_{out} = V_{DD}$
B	$V_{Tn} \le V_{in} < (V_{DD}/2)$	Linear	Saturation	$V_{out} > V_{DD}/2$
C	$Vin = V_{DD}/2$	Saturation	Saturation	$V_{out}$ drops sharply
D	$(V_{DD}/2) < V_{in} \le V_{DD}-\mid V_{Tp} \mid$	Saturation	Linear	$V_{out} < V_{DD}/2$
E	$V_{in} > V_{DD}-\mid V_{Tp} \mid$	Cutoff	Linear	$V_{out} = 0$

• Almost all the operating points are located either at the high or low output levels. The VTC of the inverter hence exhibits a very narrow transition zone (resulting from the high gain during the switching transient).

---

Day 16: CMOS Switching Threshold and Dynamic Simulations

16.1 CMOS Inverter VTC (contd.)

16.1.1 Lab: CMOS Inverter VTC - sky130 (Wp/Wn = 0.84u/0.36u, L=0.15u)

SPICE File: day3_inv_vtc_Wp084_Wn036.spice

*** Model Description ***
.param temp=27

*** Including sky130 library files **
.lib "sky130_fd_pr/models/sky130.lib.spice" tt

*** Netlist Description **
XM1 out in vdd vdd sky130_fd_pr__pfet_01v8 w=0.84 l=0.15
XM2 out in 0 0 sky130_fd_pr__nfet_01v8 w=0.36 l=0.15
Cload out 0 50fF
Vdd vdd 0 1.8V
Vin in 0 1.8V

*** Simulation Commands ***
.op
.dc Vin 0 1.8 0.01

.control
run
setplot dc1
display
meas dc Vm find V(out) when V(out)=V(in)
.endc

.end

Output:

16.1.2 Lab: CMOS Inverter Transition time - sky130 (Wp/Wn = 0.84u/0.36u, L=0.15u)

SPICE File: day3_inv_tran_Wp084_Wn036.spice

 *** Model Description ***
.param temp=27

*** Including sky130 library files ***
.lib "sky130_fd_pr/models/sky130.lib.spice" tt

*** Netlist Description ***
XM1 out in vdd vdd sky130_fd_pr__pfet_01v8 w=0.84 l=0.15
XM2 out in 0 0 sky130_fd_pr__nfet_01v8 w=0.36 l=0.15
Cload out 0 50fF
Vdd vdd 0 1.8V
Vin in 0 PULSE(0V 1.8V 0 0.1ns 0.1ns 2ns 4ns)

*** Simulation Commands ***
.tran 1n 10n

.control
run

let vdd=1.8
let slew_low_rise_thr=0.2*vdd
let slew_high_rise_thr=0.8*vdd
let slew_high_fall_thr=0.8*vdd
let slew_low_fall_thr=0.2*vdd
let tp_thr=0.5*vdd

meas tran t_rise TRIG v(out) VAL=slew_low_rise_thr RISE=1 TARG v(out) VAL=slew_high_rise_thr RISE=1
meas tran t_fall TRIG v(out) VAL=slew_high_fall_thr FALL=1 TARG v(out) VAL=slew_low_fall_thr FALL=1
meas tran t_pLH TRIG v(in) VAL=tp_thr FALL=2 TARG v(out) VAL=tp_thr RISE=2
meas tran t_pHL TRIG v(in) VAL=tp_thr RISE=2 TARG v(out) VAL=tp_thr FALL=2
.endc

.end

Output:

16.2 Evaluating the Robustness of the CMOS Inverter

16.2.1 Static Behaviour Robustness: (1) Switching Threshold, VM

• The switching threshold, $V_M$, is defined as the point where $V_{in} = V_{out}$.
• Graphically it can be found from the intersection of the VTC with the $V_{in} = V_{out}$ line.
• In the region around $V_M$, both PMOS and NMOS are in saturation, since $V_{DS} = V_{GS}$.
• An analytical expression for $V_M$ can be obtained by equating the currents through the PMOS and NMOS transistors, $I_{DSn}=I_{DSp}$.
• Depending on the supply voltage, $V_{DD}$ and the Channel length, $L$, of the devices, there can be two cases:
  1. Devices are Velocity Saturated
  2. Velocity Saturation does not occur

Note: For the following derivations, we ignore the effects of Channel Length Modulation for simplicity.

Case 1: Devices are Velocity-Saturated - $V_{DSAT}&lt;(V_M-V_{TH})$

• This case is applicable to short-channel devices or when the supply voltage is high so that the devices are in velocity saturation.

$I_{DSn} = -I_{DSp}$
$i.e.,$
$~~~~ I_{DSn} + I_{DSp} = 0$

$k_n \left[ (V_M - V_{THn})V_{DSATn} - \dfrac{V_{DSATn}^2}{2} \right] + k_p \left[ (V_M - V_{DD} - V_{THp})V_{DSATp} - \dfrac{V_{DSATp}^2}{2} \right] = 0$

$k_n V_{DSATn} \left[ V_M - V_{THn} - \dfrac{V_{DSATn}}{2} \right] + k_p V_{DSATp} \left[ V_M - V_{DD} - V_{THp} - \dfrac{V_{DSATp}}{2} \right] = 0$

$Solving for V_M:$

$\boxed{V_M = \dfrac{\left( V_{THn}+\dfrac{V_{DSATn}}{2} \right) + r \left( V_{DD}+V_{THp}+\dfrac{V_{DSATp}}{2} \right)}{1+r}},$

$where, ~ \boxed{r=\dfrac{k_p V_{DSATp}}{k_n V_{DSATn}} = \dfrac{\upsilon_{satp} W_p}{\upsilon_{satn} W_n}}$ (assuming for identical oxide thickness for PMOS and NMOS transistors)

• Now, for large values of $V_{DD}$ compared to the threshold voltages $(V_{THp}, V_{THn})$ and saturation voltages $(V_{DSATp}, V_{DSATn})$, the above equation can be approximated to:

$~~~~~~~~~~~~~~~~ \boxed{V_M \approx \dfrac{rV_{DD}}{1+r}}$

• The switching threshold is determined by the ratio, $r$ - which is a measure of the relative drive strengths of the PMOS and NMOS transistors.

• For comparable values for low and high noise margins, $V_M$ is desired to be located around the centre of the available voltage swing (or at $V_{DD}/2$ as CMOS logic has rail-to-rail swing). This implies:
  $~~~~~~~~ r \approx 1$
  $~~~~~~~~ i.e., k_p V_{DSATp} = k_n V_{DSATn}$
  $\boxed{(W/L)_p = (W/L)_n * \dfrac{{k_n}^\prime V_{DSATn}}{{k_p}^\prime V_{DSATp}}}$

• To move the $V_M$ upwards, a larger value of $r$ is needed, which in other words is to make the PMOS wider.

• On the other hand, to move the $V_M$ downwards, the NMOS must be made wider.

• If a target design value for $V_M$ is desired, we can derive the required ratio of PMOS vs. NMOS transistor sizes in a similar manner:

$~~~~~~~~ \boxed{\dfrac{(W/L)_p}{(W/L)_n} = \dfrac{k_n^\prime V_{DSATn} \left[ V_M - V_{THn} - \dfrac{V_{DSATn}}{2} \right]}{k_p^\prime V_{DSATp} \left[ V_{DD} - V_M + V_{THp} + \dfrac{V_{DSATp}}{2}\right]}}$
$~~~~~~~~ Note:$ Make sure that the assumption that both devices are velocity-saturated still holds for the chosen operation point.

Case 2: Velocity Saturation does not occur - $V_{DSAT}&gt;(V_M-V_{TH})$

• This case is applicable for long-channel devices or when the supply voltages are low, that the devices are not velocity saturated.
• Using a similar analysis done in Case 1 above, we derive the switching threshold, $V_M$ to be:

$~~~~~~~~ \boxed{V_M = \dfrac{V_{THn} + r(V_{DD} + V_{THp})}{1+r}}, ~~~~ where ~ r = \sqrt{\dfrac{-k_p}{k_n}}$

• The switching threshold, $V_M$ is relatively insensitive to variations in the device ratio.
  • Small variations of the ratio (e.g., 3 or 2.5) do not disturb the transfer characteristic that much.

---

Day 17: CMOS Noise Margin Robustness Evaluation

17.1 Static Behaviour Robustness: (2) Noise Margin

• In digital circuits, if the magnitude of the "noise voltage" at a node is too large, logic errors can be introduced into the system.
• However, if the noise amplitude is less than a specified value, called the noise margin, the noise signal will be attenuated as it passes through the logic gate or circuit, and the logic signals will be transmitted without any errors.
• i.e., Noise margin is the amount of noise that a CMOS circuit could withstand without compromising the operation of circuit.

• Noise margin makes sure that:
  • any signal which is logic 1 with finite noise added to it, is still recognized as logic 1 and not logic 0.
  • similarly, any signal which is logic 0 with finite noise added to it, is still recognized as logic 0 and not logic 1.

17.1.1 Noise Margin Definition

• Let us consider the case of two CMOS inverters connected back-to-back with one driving the next.
• The following images show an ideal, a piece-wise linear and a realistic VTC of a CMOS inverter:

• $V_{IL}$ and $V_{IH}$ (or to be more precise, $V_{IL-MAX}$ and $V_{IH-MIN}$) are defined to be the operational points of the inverter where $\dfrac{dV_{out}}{dV_{in}} = -1$. Or, from an analog design perspective, these are the points where the gain of the inverting amplifier formed by the inverter is equal to -1.

  • Any input voltage level between 0 and $V_{IL}$ will be treated as logic 0
  • Any input voltage level between $V_{IH}$ and $V_{DD}$ will be treated as logic 1
  • The point $V_{IL}$ occurs when the NMOS is biased in saturation region and the PMOS is biased in the linear region.
  • Similarly, the point $V_{IH}$ occurs when the NMOS is biased in linear region and the PMOS is biased in the saturation region.

  -----------------------
  Output Characteristics:
  -----------------------
  VOL_Min : Minimum output voltage that the logic gate can drive for a logic "0" output.
  VOL_Max : Maximum output voltage that the logic gate will drive corresponding to a logic "0" output.
  VOH_Min : Minimum output voltage that the logic gate will drive corresponding to a logic "1" output.
  VOH_Max : Maximum output voltage that the logic gate can drive for a logic "1" output.
  
  ----------------------
  Input Characteristics:
  ----------------------
  VIL_Min : The minimum input voltage to the gate corresponding to logic "0" -- is equal to the VSS
  VIL_Max : The maximum input voltage to the gate that will be recognized as logic "0"
  VIH_Min : The minimum input voltage to the gate that will be recognized as logic "1"
  VIH_Max : The maximum input voltage to the gate corresponding to logic "1" -- is equal to the VDD
  

• Obviosuly, for proper operation of the logic gate in the presence of noise:

  • $V_{OL-MAX} &lt; V_{IL-MAX}$
  • $V_{OH-MIN} &gt; V_{IH-MIN}$

• For $V_{in} \le V_{IL}$ , the inverter gain magnitude is less than unity, and the output change is minimal for a given change in the input voltage in this range.

• Similarly, for $V_{in} \ge V_{IH}$ , the output change is minimal for a given input voltage in this range, again because of the same reason that the gain magnitude is less than unity.

• However, when the input voltage is in the range $V_{IL} &lt; V_{in} &lt; V_{IH}$ , the gain magnitude is greater than one, and the output signal amplitude changes drastically.

  • This region is called the undefined range (from a digital design standpoint), since if the input voltage is inadvertently pushed into this range by a noise signal, the output may change logic state introducing an error.

• The noise margins are defined as thus defined as follows:

  • Low-level Noise Margin, $~ NM_L ~ = V_{IL-MAX} - V_{OL-MAX}$
  • High-level Noise Margin, $NM_H = V_{OH-MIN} - V_{IH-MIN}$
  • Noise Margin, $NM = Min(NM_L, NM_H)$

17.1.2 Noise Margin Robustness against variations in Device Ratio

17.1.3 Lab: Noise Margin - sky130 Inverter (Wp/Lp=1u/0.15u, Wn/Ln=0.36u/0.15u)

SPICE File: day4_inv_noisemargin_wp1_wn036.spice

*** Model Description ***
.param temp=27

*** Including sky130 library files ***
.lib "sky130_fd_pr/models/sky130.lib.spice" tt

*** Netlist Description ***
XM1 out in vdd vdd sky130_fd_pr__pfet_01v8 w=1 l=0.15
XM2 out in 0 0 sky130_fd_pr__nfet_01v8 w=0.36 l=0.15
Cload out 0 50fF
Vdd vdd 0 1.8V
Vin in 0 1.8V

*** Simulation Commands ***
.op
.dc Vin 0 1.8 0.01

.control
run
setplot dc1
display
let dVout = deriv(V(out))
meas dc vil find V(in) when dVout=-1 cross=1
meas dc vih find V(in) when dVout=-1 cross=2
meas dc voh find V(out) when dVout=-1 cross=1
meas dc vol find V(out) when dVout=-1 cross=2

let NML = vil - vol
let NMH = voh - vih
print NML
print NMH
.endc

.end

Output:

---

Day 18: CMOS Power Supply and Device Variation Robustness Evaluation

18.1 Static Behaviour Robustness: (3) Power Supply Variations/ Scaling

Here, we take a look at the effect of power supply scaling on the static behaviour of the CMOS Inverter.

18.1.1 Lab - Supply Scaling - sky130 Inverter (Wp/Wn=1u/0.36u, L=0.15u)

SPICE File: day5_inv_supplyvariation_Wp1_Wn036.spice

*** Model Description ***
.param temp=27

*** Including sky130 library files ***
.lib "sky130_fd_pr/models/sky130.lib.spice" tt

*** Netlist Description ***
XM1 out in vdd vdd sky130_fd_pr__pfet_01v8 w=1 l=0.15
XM2 out in 0 0 sky130_fd_pr__nfet_01v8 w=0.36 l=0.15
Cload out 0 50fF
Vdd vdd 0 1.8V
Vin in 0 1.8V

.control
let powersupply     = 1.8
let powersupply_min = 0.8
let increment_step  = -0.2

dowhile powersupply >= powersupply_min
    dc Vin 0 1.8 0.01
    let powersupply = powersupply + increment_step
    alter Vdd = powersupply
end

plot dc1.V(out) dc2.V(out) dc3.V(out) dc4.V(out) dc5.V(out) dc6.V(out) vs in 
+ xlabel 'Input Voltage(V)' ylabel 'Output Voltage(V)' title 'Inverter VTC for different VDD'

plot deriv(dc1.V(out)) deriv(dc2.V(out)) deriv(dc3.V(out)) deriv(dc4.V(out)) deriv(dc5.V(out)) deriv(dc6.V(out)) vs in 
+ xlabel 'Input Voltage(V)' ylabel 'Gain (dVout/dVin)' title 'Inverter Gain for different VDD'
.endc

.end

Output:

We can see from the simulations that:

• The CMOS Inverter continues to work well at even 0.8V (below half the original supply voltage of 1.8V) close to the transistor threshold voltages!
• Since the transistor ratio, $r$ is fixed, the switching threshold, $V_M$ is approximately proportional to the $V_{DD}$.
• The gain of the inverter in the transition region increases with a reduction of the supply voltage.
• The width of the transition region also reduces when the supply voltage is scaled down compared to the original $V_{DD}$.

However, given these improvements in DC characteristics, we cannot choose to operate all our digital circuits at these low supply voltages:

• Reducing the supply voltage yields a significant reduction in the energy dissipation, but it is absolutely detrimental to the performance of the gate, increasing the transition times greatly.
• The DC characteristic becomes increasingly sensitive to variations in the device parameters such as the transistor threshold, once supply voltages and intrinsic voltages become comparable.
• Scaling the supply voltage means reducing the signal swing. While this typically helps to reduce the internal noise in the system (such as caused by crosstalk), it makes the design more sensitive to external noise sources that do not scale.

18.2 Static Behaviour Robustness: (4) Device Variations

While we design a gate for nominal operation conditions and typical device parameters, the actual operating temperatures might very over a large range, and the device parameters after fabrication will deviate from the nominal values used in the design process.

The DC characteristics of the static CMOS inverter turn out to be rather insensitive to these variations, and the gate remains functional over a wide range of operating conditions.

18.2.1 Sources of Variation

1) Etching Process Variations

2) Oxide Thickness Variation

18.2.2 Lab - Device Variations - sky130 Inverter - (Wp:7u --> 0.42u, Wn:0.36u --> 7u, L=0.15u)

In this exercise, we try to capture the CMOS Inverter's robustness in the case of a ridiculously extreme case of device width variation.

SPICE File: day5_inv_devicevariation_wp7_wn042.spice

*** Model Description **
.param temp=27
.param Wp=0.84
.param Wn=0.42

*** Including sky130 library files ***
.lib "sky130_fd_pr/models/sky130.lib.spice" tt

*** Netlist Description ***
XM1 out in vdd vdd sky130_fd_pr__pfet_01v8 w={Wp} l=0.15
XM2 out in 0 0 sky130_fd_pr__nfet_01v8 w={Wn} l=0.15
Cload out 0 50fF
Vdd vdd 0 1.8V
Vin in 0 1.8V

*** Simulation Commands ***
.dc Vin 0 1.8 0.01

.control
alterparam Wp=0.84
alterparam Wn=0.42
reset
run

alterparam Wp=0.42
alterparam Wn=7
reset
run

alterparam Wp=7
alterparam Wn=0.36
reset
run

plot dc1.V(out) dc2.V(out) dc3.V(out) vs in 
+ xlabel 'Input Voltage(V)' ylabel 'Output Voltage(V)' title 'Inverter VTC for Extreme Device Variations'
.endc
.end

From the simulation results:

1. The shift in $V_M$ for such a large change in device variations is relatively very small.
2. The variation in the $NM_H$ and $NM_L$ is also very small.

Output:

---

Day 19 - PVT Corner Analysis (Post-Synthesis Timing) of the RISC-V CPU Design

The STA checks are performed across all the corners to confirm the design meets the target timing requirements.

• The worst max path (Setup-critical) corners in the sub-40nm process nodes are usually: ss_LowTemp_LowVolt, ss_HighTemp_LowVolt (Slowest corners)
• The worst min path (Hold-critical) corners being: ff_LowTemp_HighVolt,ff_HighTemp_HighVolt (Fastest corners)
• The below tcl script sta_across_pvt.tcl can be run to performt the STA across the PVT corners for which the sky130 lib files are available:

  set list_of_lib_files(1) "sky130_fd_sc_hd__tt_025C_1v80.lib"
  set list_of_lib_files(2) "sky130_fd_sc_hd__ff_100C_1v65.lib"
  set list_of_lib_files(3) "sky130_fd_sc_hd__ff_100C_1v95.lib"
  set list_of_lib_files(4) "sky130_fd_sc_hd__ff_n40C_1v56.lib"
  set list_of_lib_files(5) "sky130_fd_sc_hd__ff_n40C_1v65.lib"
  set list_of_lib_files(6) "sky130_fd_sc_hd__ff_n40C_1v76.lib"
  set list_of_lib_files(7) "sky130_fd_sc_hd__ss_100C_1v40.lib"
  set list_of_lib_files(8) "sky130_fd_sc_hd__ss_100C_1v60.lib"
  set list_of_lib_files(9) "sky130_fd_sc_hd__ss_n40C_1v28.lib"
  set list_of_lib_files(10) "sky130_fd_sc_hd__ss_n40C_1v35.lib"
  set list_of_lib_files(11) "sky130_fd_sc_hd__ss_n40C_1v40.lib"
  set list_of_lib_files(12) "sky130_fd_sc_hd__ss_n40C_1v44.lib"
  set list_of_lib_files(13) "sky130_fd_sc_hd__ss_n40C_1v76.lib"
  
  for {set i 1} {$i <= [array size list_of_lib_files]} {incr i} {
  read_liberty ./timing_libs/$list_of_lib_files($i)
  read_verilog ./riscv_pipelined_Final_netlist.v
  link_design riscv_core
  current_design
  read_sdc riscv_core_synthesis.sdc
  check_setup -verbose
  report_checks -path_delay min_max -fields {nets cap slew input_pins fanout} -digits {4} > ./sta_output/min_max_$list_of_lib_files($i).txt
  
  exec echo "$list_of_lib_files($i)" >> ./sta_output/sta_worst_max_slack.txt
  report_worst_slack -max -digits {4} >> ./sta_output/sta_worst_max_slack.txt
  
  exec echo "$list_of_lib_files($i)" >> ./sta_output/sta_worst_min_slack.txt
  report_worst_slack -min -digits {4} >> ./sta_output/sta_worst_min_slack.txt
  
  exec echo "$list_of_lib_files($i)" >> ./sta_output/sta_tns.txt
  report_tns -digits {4} >> ./sta_output/sta_tns.txt
  
  exec echo "$list_of_lib_files($i)" >> ./sta_output/sta_wns.txt
  report_wns -digits {4} >> ./sta_output/sta_wns.txt
  }
  

---

Days 20-24: Advanced Physical Design using OpenLANE/ Sky130

Day 20: Inception of open-source EDA, OpenLANE and Sky130 PDK

• IC terminologies
  • Chip Package
  • Pads & Padring
  • Core, die
  • IPs, Macros

20.1 Overview of ASIC Design Flow using OpenLane

• ASIC implementation consists of numerous steps involving lots of detailed sub-processes at each step.
• A design methodology is needed for a successful ASIC implementation without any hiccups.
• The methodology is implemented through a flow that pieces together different tools to carry out the different steps of the design process from RTL to GDSII tapeout.

20.1.1 Simplified RTL to GDSII ASIC Design Flow

1. Synthesis: Converts RTL to a circuit using components from the Standard Cell Library (SCL)
2. Floor/ Power-planning: Floor-planning
   • Chip Floor-planning: Partition chip die area between the different building macros, IPs and place the I/O pads in the pad ring
   • Macro Floor-planning: Define the dimensions, pin locations and the internal row definitions of the macro Power-Planning: Design the Power Distribution Network (PDN) to ensure stable and efficient power delivery taking into account the different voltage rails, power usage trends and requirements of different blocks within the digital core, I/O and other AMS circuits.
3. Placement: Place the cells on the floorplan rows aligned with the sites, non-overlapping with each other.
   Usually done in 2 dteps: Global and Detailed
4. Clock Tree Synthesis (CTS): Create a balanced clock distribution network that ensure minial skew and achieves proper timing across the entire design.
   Different types: H-tree, X-tree, Fish bone, Clock mesh.
5. Routing: Implement the interconnections using the available metal layers.
   • Metal traacks form a routing grid, which is huge
   • Perform routing using a divide and conquer approach:
     • Global routing generates routing guides
     • Detailed routing uses the routing guides to implement the actual wires
6. Sign-Off:
   • Physical Verification
     • Design Rule Checking (RDC)
     • Layout vs. Schematic (LVS)
   • Timing Verification
     • Static Timing Analysis (STA)

20.1.2 OpenLANE ASIC Design Flow

Main requirements of Digital ASIC Design:

• RTL Design
• EDA Tools
• PDK

Open Source RTL IPs and competitive EDA tools have been available.
However, an OpenSource PDK was not available until Google collaborated with SkyWater to open source the skywater-130nm PDK.

Process Design Kit (PDK) : Interface between FAB and the designer, Process Design Kit, include:

• Process Design Rule: DRC, LVS, PEX
• Device Models
• Digital Standard Cell Libraries
• I/O Libraries

Introduction to OpenLANE

• Started as an Open-Source Flow for a True Open Source Tape-out Experiment
• Main Goal: Produce a clean GDSII with no human intervention (no-human-in-the-loop)
• Clean means: No LVS Errors and No DRC Errors
• Tuned for SkyWater 130nm Open PDK. Also supports XFAB180 and GF130G
• Containerized: Functional out of the box
• Can be used to harden Macros and Chips
• Two modes of operation:
  • Autonomous or Interactive
• Supports Design Space Exploration: Find the best set of flow configurations

OpenLANE ASIC Flow

• OpenLANE project GitHub Page: OpenLANE
• OpenLANE ReadMe

1. Synthesis
   • Yosys: RTL Synthesis
   • abc: Technology mapping
2. Post-synthesis STA
   • OpenSTA
3. DFS Insertion
   • Fault: Scan Insertion, ATPG, Test Patterns compaction, Fault coverage, Fault simulation
4. Floor-planning, Placement, CTS, Fake Antenna Diodes Insertion, Global Routing
   • OpenROAD: Also called automated PnR
5. Logical Equivalence Check (LEC)
   • Yosys
     • Every time the netlist is modified (ECO), verification must be performed
       • CTS modifies the netlist
       • Post Placement optimizations modifies the netlist
     • LEC is used to formally confirm that the function did not change by modifying the netlist
6. Detailed Routing
   • TritonRoute
7. Fake Antenna Diodes removal
   • Custom scripts in OpenROAD
8. RC Extraction:
   • DEF2SPEF
9. Post-route STA
   • OpenSTA
10. Physical Verification
    • Magic is used for Design Rules Checking and SPICE Extraction from Layout
    • Netgen is used for LVS
      • Extracted SPICE by Magic vs. Verilog netlist
11. GDSII streaming
    • Magic

OpenROAD

• Everything in Floorplanning through Routing in the OpenLANE flow is done using OpenROAD and its various sub-utilities.
• Also called automated PnR (Place and Route)
• Performs:
  • Floor/Power Planning
  • End Decoupling Capacitors and Tap cells insertion
  • Placement: Global and Detailed
  • Post placement optimization
  • Clock Tree Synthesis (CTS)
  • Routing: Global and Detailed

Handling of Antenna Rules Violations in OpenLANE flow

• When a metal wire segment is fabricated, it can act as an antenna.
  • Reactive ion etching causes charge to accumulate on the wire.
  • Transistor gates can be damaged during fabrication
    
• Two solutions:
  
  • Bridging attaches a higher layer intermediary
    • Requires Router awareness (not there yet!)
  • Add antenna diode cell to leak away charges
    • Antenna diodes are provided by the SCL
• Methodology followed: a preventive approach
  • Add a Fake Antenna Diode next to every cell input after placement
  • Run the Antenna Checker (Magic) on the routed layout
  • If the checker reports a violation on the cell input pin, replace the Fake Diode cell by a real one

Lab: Familiarize with OpenLANE flow

Objectives:
Using an existing design provided in the OpenLANE package to:

• Familiarize with the OpenLANE directory structure and different input files
• Familiarize with the OpenLANE flow
• Analyse the intermediate step results
• Learn about the different control knobs and switches available for design space exploration
  The OpenLANE flow can be configured using the following available variables for design
  • Flow Configuration variables: https://openlane.readthedocs.io/en/latest/reference/configuration.html
  • PDK Configuration variables: https://openlane.readthedocs.io/en/latest/reference/pdk_configuration.html

Design used for this exercise: picorv32a

1. To invoke OpenLANE, cd to the home directory of OpenLANE and run docker:
   docker run -it -v $(pwd):/openLANE_flow -v $PDK_ROOT:$PDK_ROOT -e PDK_ROOT=$PDK_ROOT -u $(id -u $USER):$(id -g $USER) efabless/openlane:v0.21
   where the env variable PDK_ROOT points to the directory path containing the sky130A library.

2. The entry point for OpenLANE is the ./flow.tcl script. This script is used to run the flow, start interactive sessions, select the configuration and create OpenLane design files.

   • To run the automated flow for a design:

     ./flow.tcl -design <design_name>
     

   • To start an interactive session:

     ./flow.tcl -interactive
     

3. We will be using the interactive mode to learn about the different steps in the flow.

   • The commands to start an interactive session and run the synthesis of the picorv32a example design are given below:

     ./flow.tcl -interactive
     package require openlane 0.9
     prep -design picorv32a
     run_synthesis
     

• Synthesis Result:

   === picorv32a ===
   
      Number of wires:              14596
      Number of wire bits:          14978
      Number of public wires:        1565
      Number of public wire bits:    1947
      Number of memories:               0
      Number of memory bits:            0
      Number of processes:              0
      Number of cells:              14876
        sky130_fd_sc_hd__a2111o_2       1
        sky130_fd_sc_hd__a211o_2       35
        sky130_fd_sc_hd__a211oi_2      60
        sky130_fd_sc_hd__a21bo_2      149
        sky130_fd_sc_hd__a21boi_2       8
        sky130_fd_sc_hd__a21o_2        57
        sky130_fd_sc_hd__a21oi_2      244
        sky130_fd_sc_hd__a221o_2       86
        sky130_fd_sc_hd__a22o_2      1013
        sky130_fd_sc_hd__a2bb2o_2    1748
        sky130_fd_sc_hd__a2bb2oi_2     81
        sky130_fd_sc_hd__a311o_2        2
        sky130_fd_sc_hd__a31o_2        49
        sky130_fd_sc_hd__a31oi_2        7
        sky130_fd_sc_hd__a32o_2        46
        sky130_fd_sc_hd__a41o_2         1
        sky130_fd_sc_hd__and2_2       157
        sky130_fd_sc_hd__and3_2        58
        sky130_fd_sc_hd__and4_2       345
        sky130_fd_sc_hd__and4b_2        1
        sky130_fd_sc_hd__buf_1       1656
        sky130_fd_sc_hd__buf_2          8
        sky130_fd_sc_hd__conb_1        42
        sky130_fd_sc_hd__dfxtp_2     1613
        sky130_fd_sc_hd__inv_2       1615
        sky130_fd_sc_hd__mux2_1      1224
        sky130_fd_sc_hd__mux2_2         2
        sky130_fd_sc_hd__mux4_1       221
        sky130_fd_sc_hd__nand2_2       78
        sky130_fd_sc_hd__nor2_2       524
        sky130_fd_sc_hd__nor2b_2        1
        sky130_fd_sc_hd__nor3_2        42
        sky130_fd_sc_hd__nor4_2         1
        sky130_fd_sc_hd__o2111a_2       2
        sky130_fd_sc_hd__o211a_2       69
        sky130_fd_sc_hd__o211ai_2       6
        sky130_fd_sc_hd__o21a_2        54
        sky130_fd_sc_hd__o21ai_2      141
        sky130_fd_sc_hd__o21ba_2      209
        sky130_fd_sc_hd__o21bai_2       1
        sky130_fd_sc_hd__o221a_2      204
        sky130_fd_sc_hd__o221ai_2       7
        sky130_fd_sc_hd__o22a_2      1312
        sky130_fd_sc_hd__o22ai_2       59
        sky130_fd_sc_hd__o2bb2a_2     119
        sky130_fd_sc_hd__o2bb2ai_2     92
        sky130_fd_sc_hd__o311a_2        8
        sky130_fd_sc_hd__o31a_2        19
        sky130_fd_sc_hd__o31ai_2        1
        sky130_fd_sc_hd__o32a_2       109
        sky130_fd_sc_hd__o41a_2         2
        sky130_fd_sc_hd__or2_2       1088
        sky130_fd_sc_hd__or2b_2        25
        sky130_fd_sc_hd__or3_2         68
        sky130_fd_sc_hd__or3b_2         5
        sky130_fd_sc_hd__or4_2         93
        sky130_fd_sc_hd__or4b_2         6
        sky130_fd_sc_hd__or4bb_2        2
   
      Chip area for module '\picorv32a': 147712.918400

• Flop Ratio

  Flop Ratio = (Total no. of Flops/ Total no. of cells in the design)
             = 1613/ 14876
             = 10.843%
  

• NOTE: The order of precedence of the config files in the OpenLANE flow is as follows, with the settings in the highest priority config overriding the values set in the previous config files:
  From lowest to highest:

  • Default OpenLANE config values
  • openlane/designs//config.tcl
  • openlane/designs//sky130A_sky130_fd_sc_hd_config.tcl

---

Day 21: Floorplan considerations, Placement, Library Cells

21.1 Floorplan considerations

1. Utilization factor and aspect ratio

   • Define W, H of core and die
     • Utilization Factor = (Area occupied by netlist)/(Total area of the core)
     • Usually we aim for 50-60 % Utilization Factor
   • Aspect Ratio = Height/ Width

2. Define locations of pre-placed cells (macros and IPs ?)

   • IPs/ blocks have user-defined locations and hence placed before automated PnR and are called as pre-placed cells
   • Automated PnR tools places the remaining logical cells in the design onto the chip

3. Decaps

   • Decouples the circuit from the VDD rail
   • Reduce Zpdn for the required frequencies of operation
   • Serve as a charge reservoir for the switching current demands that the VDD rail cannot satisfy.
   • Surround pre-placed cells with Decaps to compensate for the switching current demands (di/dt)

4. Power Planning

   • SSN
     • L*di/dt
       • Discharging : Ground bounce
       • Charging : Voltage Droop
     • Solution: Reduce the Vdd/ Vss parasitics ->
       • Power grid
       • Multiple VDD, VSS pins/ balls

5. Pin Placement

   • Usually: East -> West, North -> South, {East, North} -> {West, South}
   • Pin ordering is random (unless we specify explicitly ?)
   • Front-End to Back-End team communication/ handshaking needed for optimal pin placement
   • CLK ports/ pins are usually bigger to reduce the clk net resistance

6. Logical Cell placement blockage - so that no cells are placed by the PnR tool inside the IP blocks/ macro area.

Now Floorplan is ready for PnR

Lab: Run floorplan using OpenLANE and review the layout in Magic

• To run the floorplan creation, execute the following command from the OpenLANE shell: run_floorplan

• To view the floor plan in Magic:

  1) cd to the floorplan results directory for the current run: openlane/designs/<design-name>/runs/<time-stamp>/results/floorplan
  2) magic -T $PDK_ROOT/sky130A/libs.tech/magic/sky130A.tech lef read ../../tmp/merged.lef def read ./picorv32a.floorplan.def &
  

Floorplan view in Magic (FP_IO_MODE=1, Random equidistant mode)
Floorplan Zoomed in at (0,0)
FP_IO_MODE=0 (Matching mode)
FP_IO_MODE=1 Zoomed

21.2 Placement and Routing

1. Bind netlist with physical cells

   • Library files
     • Shape, dimension info, power & timing/ delay info
     • Various flavors of all available std cells

2. Placement

   • The location of pre-placed cells are fixed and PnR tools will not place any cells inside that area.
   • Initial global placement based on the input, output pins to reduce wire lengths

3. Optimize placement

   • Ensure that Signal Integrity is maintained
     • Estimate the wire length and capacitance and insert repeaters/ buffers
     • MaxTrans delay/ signal slew

4. Congestion aware placement using RePlAce followed by detailed placement using OpenDP

   • Global placement: HPWL (Half-Parameter Wire Length) based

Lab: Run placement

• The run_placement command runs the global placement followed by detailed placement.
• First the global placement happens, where the main objective is to reduce the wire length. Algorithm used is Half-Parameter Wire Length (HPWL).
• Then detailed placement is performed to legalize the globally placed components.

run_placement
Layout after placement
Layout after placement (Zoomed)

21.3 Cell Design Flow

Library

• Std cells of different:
  • logic functionality
  • size/ drive strengths
  • threshold voltages

Stages

1. Inputs for Cell Design Flow
   From foundry, PDKs:

   • DRC, LVS rules (eg: lambda-based design rules)
   • SPICE modelsfor the NMOS & PMOS devices
   • Library and user-defined specifications
     • Cell height (separation b/w VDD and GND rails) must be maintained
     • Cell width is determined by the drive strengths required
     • Supply voltage
     • Metal layers requirements
     • Pin locations
     • Drawn gate-length

2. Design Steps

   • Circuit Design

     • Transistor sizing based on current drive, voltage transfer and other performance/ quality parameter requirements

   • Layout Design

     • Art of layout - Euler's path and stick diagram

   • Characterization

     1. Read the nmos, pmos models
     2. Read the extracted spice netlist
     3. Recognize the behaviour of the cell
     4. Read the sub-circuits
     5. Attach the power sources
     6. Apply the stimulus
     7. Provide the necessary output capacitance
     8. Provide the necessary simulation commands

   • Now, feed in all this to GUNA characterization tool --> to generate Timing, noise, power .libs, function

3. Outputs

   • CDL (Circuit Description Language)
   • GDSII, LEF (dimensions), extracted spice netlist (.cir)
   • Timing, noise, power .libs, function

Timing characterization parameters

1. Timing threshold definitions

   Parameter	Usually used values
   slew_low_rise_thr	10%, 20%
   slew_high_rise_thr	90%, 80%
   slew_low_fall_thr	10%, 20%
   slew_high_fall_thr	90%, 80%
   in_rise_thr	50%
   in_fall_thr	50%
   out_rise_thr	50%
   out_fall_thr	50%

2. Propagation delay

   • propagation delay = time(out_thr) - time(in_thr)
     • Problematic cases:
       • Choice of threshold levels
       • Large RC delays (due to improper designs and/ or large loads, long routes)

3. Transition time

   • t_rise = time(slew_high_rise_thr) - time(slew_low_rise_thr)
   • t_fall = time(slew_low_fall_thr) - time(slew_high_fall_thr)

---

Day 22: Library Cell design using Magic and Characterization using Ngspice

22.1 16-Mask CMOS Process

TODO: Documentation

Lab: Introduction to Sky130 basic layers layout and LEF using inverter

• Clone a custom standard cell design from the following github repo for this exercise https://github.com/nickson-jose/vsdstdcelldesign.git
• To open the design in magic: magic -T sky130A.tech sky130_inv.mag
• Get familiarized with the different layers in sky130 technology.
• To get the details about any drawn element in the layout, hover the mouse pointer over it and press s to select it (pressing multiple times selects the elements hierarchically). Then, from the tkcon shell, use the command what to print the details:

sky130 Layers in Magic for an Inverter

Lab: Create the Inverter Standard Cell layout and extract the SPICE netlist

• The steps to layout a custom inverter standard cell in Magic is explained in this github repo: vsdstdcelldesign
  • Magic has an interactive DRC engine - DRC violations are updated continuously in Magic every time we make a change (draw, erase, move) in the layout.
  • When we make small changes to an existing layout, we can find out immediately if we have introduced errors, without having to completely recheck the entire layout.
• To extract the SPICE netlist of the layout including the parasitics, switch to the tkcon shell of Magic:

  extract all
  ext2spice cthresh 0 rthresh 0
  ext2spice
  

• Open the extracted netlist and correct the following:
  • Ensure the Scaling: Open the layout in magic, enable the grid (Window menu -> Grid on), and zoom in to select one unit box. Find its dimensions by the command box from the tkcon shell
  • Include the SPICE models for sky130 short-channel PMOS and NMOS.
  • Change the PMOS and NMOS model names to match the ones in the included model files - pshort_model.0, nshort_model.0.

Lab: Create a SPICE deck to run a simple transient simulation using ngspice

• Modify the spice file to run a sample transient simulation using ngspice:

  • Add VDD and GND:

    VDD VPWR 0 3.3V
    VSS VGND 0 0V
    

  • Add a pulse source to the input node: Va A VGND PULSE(0V 3.3V 0 0.1ns 0.1ns 2ns 4ns)
  • Transient simulation: .tran 1n 20n
  • Finally, for some weird reasons, ngspice throws an unknown subckt error with transistor instance names starting with X. So, modify the instance names to M0 and M1

  SPICE deck to run trans sim using the extracted netlist
  Trans sim results with Waveforms

Lab: Introduction to DRC using Magic tool

• Obtain the tutorial files for DRC labs from the following link:

  wget http://opencircuitdesign.com/open_pdks/archive/drc_tests.tgz
  tar xfz drc_tests.tgz
  

• The Design Rules for Skywater 130nm technology can be found here: https://skywater-pdk.readthedocs.io/en/main/rules.html

DRC Lab 1: met3.mag

• To open Magic using OpenGL or Cairo graphical interfaces, invoke magic using the -d option:

  • For OpenGL: magic -d XR &
  • For Cairo: magic -d OGL &

• Open the met3.mag tutorial file in Magic either via the command line or from the GUI via File -> Open...

• To view the DRC errors/ violations flagged for an area:

  • Position the cursor box around the required area by using the left and right mouse buttons.
  • Commands to the tkcon console can be passed without leaving the main layout GUI window by pressing the : key followed by entering the required command
  • For example, to view the DRC error for the m3.2 section, position the cursor box around it and type: :drc why
  • The console window will now display the DRC rule that is being violated

  Rule M3.2: Spacing of metal 3 to metal 3 - 0.300µm

• Vias are a kind of derived layer in Magic, in which the drawn via represents an area that is filled with contact cuts. The contact cuts (which is essentially the Mask layer for VIA2 in the output GDS) don't actually exist in the drawn layout view. They are created from rules in the CIF output section of the tech file that tell Magic how to draw contact cuts in the drawn contact area.

  • Draw an area of M3 contact by enclosing some area in the cursor box using the left and right mouse buttons, hover the mouse pointer over the m3contact icon in the sidebar and click the middle mouse-button (or, press the p key)
  • Otherwise, use the paint m3contact command after enclosing the required area in the cursor box.
  • To view the M3 contact cuts, from the layout window, type :cif see VIA2 This view in the layout window is called a feedback entry and can be dismissed using the feedback clear command.
  • As a sidenote, rules like these will always be correct by design and can be confirmed by measuring the distance from the contact cut to the edge of M3 contact by drawing a cursor box.
    • To align the cursor box to the edge of the via shown in the CIF view, use the snap int command.

  Rule M3.4: Via2 must be enclosed by Met3 by at least 0.065µm

DRC Lab 2: poly.mag - Exercise to fix poly.9 error in Sky130 tech-file

• https://skywater-pdk.readthedocs.io/en/main/rules/periphery.html#poly

• poly.9: Poly resistor spacing to poly or spacing (no overlap) to diff/tap 0.480 µm

• This exercise deals with fixing an incomplete DRC rule in the sky130A.tech file

• The section shown below is violating the poly.9 DRC rule, but it is not reported as a DRC violation due to the rule being incompletely implemented in the sky130A.tech file

  Rule poly.9: Poly resistor spacing to poly or spacing (no overlap) to diff/tap 0.480µm

• In the sky130A.tech file:

  • The rules for poly resistor spacing to alldiffusion and nsd (nsubstratediff or N-tap) are implemented. So we need to implement the missing poly resistor spacing to poly rules.

  sky130A.tech

• • If we look at the aliases, we can see that there is a definition for allpolyres as follows: allpolyres mrp1,xhrpoly,uhrpoly,rmp

  • So we can implement the missing poly.9 spacing rules by adding the following lines to the tech file at the appropriate sections:

    spacing npres allpolynonres 480 touching_illegal \
            "poly.resistor spacing to poly < %d (poly.9)"
    
    spacing xhrpoly,uhrpoly,xpc allpolynonres 480 touching_illegal \
          "xhrpoly/uhrpoly resistor spacing to poly < %d (poly.9)"
    

• Now, from the console window, reload the tech file: tech load sky130A.tech

• The DRC checks needs to be run again, by executing: drc check

  Magic DRC engine now shows the poly resistor to poly spacing error

DRC Lab 3: poly.mag - Exercise to implement poly resistor spacing to diff and tap

• The additions we made for the poly.9 DRC rule are still not complete. We can check this by creating two copies of the three resistors (npolyres, ppolyres and xhrpolyres)

• We will add ndiffussion, pdiffusion, nsubstratendiff & psubstratepdiff around the two copies of the three poly resistors as shown.

• Also draw an nwell under the pdiffusion and N-tap (nsubstratendiff) to avoid the flagging of any diffusion-related DRC errors since we are not interested in them for this exercise.

• From the layout, we can see that all the poly resistors except the npolyres are showing the DRC spacing violations. The npolyres is only flagging the DRC spacing violation to the N-tap. This can be fixed by changing the npres spacing rule to consider all diffusion instead of just nsd

Before	After
spacing npres *nsd 480 touching_illegal \ "poly.resistor spacing to N-tap < %d (poly.9)"	spacing npres alldiff 480 touching_illegal \ "poly.resistor spacing to N-tap < %d (poly.9)"

DRC Lab 4: nwell.mag - Challenge exercise to describe DRC error as geometrical construct

• https://skywater-pdk.readthedocs.io/en/main/rules/periphery.html#nwell
• nwell.5: Deep nwell must be enclosed by nwell by atleast… 0.400µm.
  Exempted inside UHVI or areaid.lw Nwells can merge over deep nwell if spacing too small (as in rule nwell.2)
• nwell.6: Min enclosure of nwell hole by deep nwell outside UHVI 1.030µm
• Relevant DRC rules in sky130A.tech file

• Everything in the cifoutput DRC style is implemented as a templayer and not an actual layer.

• templayer nwell_missing dnwell:

  • Depends on another templayer, dnwell_shrink
  • dnwell_shrink is defined above it.

    templayer dnwell_shrink dnwell
    shrink 1030
    

    • It takes the deep nwell and shrinks it by 1030nm (or 1.03um), which is the mininum required distance the nwell must overlap the deep nwell on the inside. What this represents is the largest open area inside the area of the drawn deep nwell.

• Coming back to the nwell_missing templayer, it starts with the dnwell and grows it by 400nm (or 0.4um, grow 400), which is the minimum required distance the deep nwell must be enclosed by the nwell. This will give the smallest nwell area needed to cover the deep nwell.

• and-not dnwell_shrink: Does nwell_missing AND-NOT dnwell_shrink, which generates a ring-shaped area that requires the least nwell to satisfy both the nwell to deep nwell inside overlap and the outside surround of deep nwell by nwell.

  • i.e., if any of this area does not contain nwell, it will be flagged as an error as either of the two rules is being violated.

• and-not nwell: Results in anything leftover is passed back to the cifmaxwidth rule in the DNWELL

• Put the cursor box around the nwell.6 drawing and execute the following commands step by step reviewing the DRC errors flagged at each step:

  cif ostyle drc   --> to see the layers of the cifoutput style drc
  cif see dnwell_shrink
  feed clear
  cif see nwell_missing
  

nwell.6 drawing	cif ostyle drc cif see dnwell_shrink
feed clear cif see nwell_missing

• NOTE:
  • Any edge based rules could be implemented using cifoutput operators but generating these layers is highly compute-intensive.
  • Hence to avoid Magic getting sluggish by these geometrically defined rules it is better to use the simple DRC edge-based rules whenever possible and put the cifoutput rules into a separate style variant, that can be run on-demand and can be prevented from running during interactive layout.
  • In sky130A.tech file, there are two variants of DRC rule styles:
    • drc fast: intended for working on backend metal layers and large synthesized digital designs without checking all the layers below metal
    • drc full: will check everything. As long as the layout is relatively small, it can be enabled during interactive layout without everything turning sluggish.
    • Switch between the two using: drc style drc(fast), drc style drc(full)

DRC Lab 5: nwell.mag - Challenge to find missing or incorrect rules and fix them

• nwell.4: All n-wells will contain metal-contacted tap (rule checks only for licon on tap) . Rule exempted from high voltage cells inside UHVI.
• Every nwell must have an n-tap layer contact inside it, which is called nsubstratencontact or nsc.
• Since there is no distance/ spacing associated with this rule, it is not possible to write this as an edge-based DRC rule. But it can be written as a cifoutput rule.
  • To the NWELL rules section, add the following cifoutput rule:

     variant (full)   # -> Run this rule only for drc(fast) style
     cifmaxwidth nwell_untapped 0 bend_illegal \
             "N-well missing tap (nwell.4))"
     variant *        # -> Run the rules below for all styles, unless explicitly specified
    

  • Now, in the style drc section, define the following templayers:

    templayer nwell_tapped           # Create a templayer nwell_tapped
    bloat-all nsc nwell              # that takes all nsc layers and expand it to fill any nwell underneath it
    templayer nwell_untapped nwell   # Now, create a second templayer nwell_untapped that starts with allnwell geometries
    and-not nwell_tapped             # and subtract all nwell_tapped geometries
    

untapped nwell being flagged for DRC violn.	tapped nwell showing no DRC violn.

---

Day 23: Pre-layout timing analysis and importance of good clock tree

Lab: Steps to convert grid info to track info

• Objective: Extract LEF file for the sky130_inv.mag and plug this custom std cell into OpenLANE flow

• From PnR point of view, we only need the following information: PnR boundary of the standard cell, the power and ground rails, and finally the input & output ports. LEF file has all these information.

• Few guidelines to be followed by the layout:

  • Input and Output ports must lie on the intersection of the vertical and horizontal tracks
  • Width of the standard cell must be an odd multiple of the track pitch
  • Height of the standard cell must be a multiple of the track vertical pitch
    • Tracks:
      • Track information is used during the routing stage as the routes can go only over the tracks.
      • Track information for the locali and the differet metal layers is specified in the pdks/sky130A/libs.tech/openlane/sky130_fd_sc_hd/tracks.info file.

• Before extracting the LEF file, we need to define the ports and set the correct class and use attributes to each port.

  1. For each layer (to be turned into port), make a box on that particular layer and input a label name along with a sticky label of the layer name with which the port needs to be associated. Ensure the Port enable checkbox is checked and default checkbox is unchecked.

  2. Next, the port class and port use attributes need to be set. These attributes are used by the LEF and DEF format read and write routines, and match the LEF/DEF CLASS and USE properties for macro cell pins.

     • Valid port classes: default, input, output, tristate, bidirectional, inout, feedthrough, and feedthru.
     • Valid port uses: default, analog, signal, digital, power, ground, and clock.
       These attributes can be set in the console window after selecting each port in the layout window.

     # For ports A, Y:
     port class input
     port use signal
     
     # For VPWR, VGND
     port class inout
     port use power
     

• Now the LEF file can be written from the console window by the following command: lef write

Lab: Adding the extracted LEF file into OpenLANE flow for picorv32a design

• Copy the LEF file into the openlane/designs/picorv32a/src directory.
• Now we need to have the cell characterized and added to the technology library.
  • This particular cell - sky130_vsdinv - it has already been characterized and added to the sky130_fd_sc_hd__*.lib files.
  • Copy the downloaded timing lib files with the characterization info for sky130_vsdinv into the openlane/designs/picorv32a/src directory.
• Modify the config.tcl with the following:
  • Set the PDK configuration variables to use the newly copied timing lib files for synthesis and sta.
  • Add the LEF file of the inverter macro using the EXTRA_LEF flow configuration variable.

    set ::env(LIB_SYNTH) $::env(OPENLANE_ROOT)/designs/picorv32a/src/sky130_fd_sc_hd__typical.lib
    set ::env(LIB_SLOWEST) $::env(OPENLANE_ROOT)/designs/picorv32a/src/sky130_fd_sc_hd__slow.lib
    set ::env(LIB_FASTEST) $::env(OPENLANE_ROOT)/designs/picorv32a/src/sky130_fd_sc_hd__fast.lib
    set ::env(LIB_TYPICAL) $::env(OPENLANE_ROOT)/designs/picorv32a/src/sky130_fd_sc_hd__typical.lib
    
    set ::env(EXTRA_LEFS) [glob $::env(OPENLANE_ROOT)/designs/$::env(DESIGN_NAME)/src/*.lef]
    

• Now invoke OpenLANE in interactive mode and after the prep -design picorv32a command is executed, run the following two commands to merge the custom cell(s)' LEF file(s) to the existing processed LEF files.

  lefs [glob $::env(DESIGN_DIR)/src/*.lef]
  add_lefs -src $lefs
  

* Synthesis result shows the sky130_vsdinv being instanced 1537 times.
Layout showing sky130_vsdinv after placement stage

• Note:
  • The run_floorplan command gave an error saying Cannot find any macros in the design after including the sky130_vsdinv cell.
  • The error message was popping up when the function basic_macro_placement was being called by run_floorplan which inturn was calling openroad ./scripts/openroad/or_basic_mpl.tcl.
  • Within or_basic_mpl.tcl, the error was being spewed out by macro_placement command and we don't have anymore info on why this error was being shown.
  • Hence, until things are cleared, run_floorplan is run manually instead with the following sub-routine calls:

    run_synthesis
    
    # run_floorplan
    init_floorplan
    place_io
    apply_def_template    # harmless I believe
    global_placement_or
    tap_decap_or
    scrot_klayout -layout $::env(CURRENT_DEF) # again, harmless
    
    # run_placement
    global_placement_or
    run_openPhySyn    # Gets skipped, so just keep it here
    run_resizer_design
    detailed_placement_or
    

23.1 Introduction to Delay Tables

• Gate/ Cell delay is a function of the input transition (slew) time and the output load capacitance, Cload
• Cell delay is calculated using Non-Linear Delay Models (NLDM). NLDM is highly accurate as it is derived from SPICE characterizations. The delay is a function of the input transition time (i.e. slew) of the cell, the wire capacitance and the pin capacitance of the driven cells.
  • A slow input transition time will slow the rate at which the cell’s transistors can change state from logic 1 to logic 0 (or logic 0 to logic 1) thereby increasing the delay of the logic gate.
  • A large output load Cload (Cnet + Cpin) also has a similar effect wherein the delay increases with the output load.
• Similar to the NLDM table for cell delay, there is an corresponding table in the library to calculate output transition as well.
• Table models are usually two-dimensional to allow lookups based on the input slew and the output load (Cload).

* Delay table for sky130_vsdinv @tt corner

• • index_1: Input slew
  • index_2: Output capacitance
  • values: Cell delay values
• Case 1: Input transition and output load values match with table index values
  • Corresponding delay value is directly picked up from the delay “values” table.
• Case 2: Output load values doesn't match with table index values
  • Interpolation is performed using the nearest available table indices to calculate the approximate delay value.

Example Interpolation algorithm in NLDM Delay table Ref: STA for Nanometer Designs - J. Bhasker, Rakesh Chadha

Lab: Configure synthesis settings to fix the timing violations and improve slack

• The synthesis results with the present settings has a huge wns slack of -26.53ns and tns of -3232.44. To obtain timing closure in post-route STA, this negative slack needs to be reduced in synthesis.

• Read back the Synthesis configuration variables that could be potentitally wrecking the timing:

  • echo $::env(SYNTH_STRATEGY) :== AREA 0
  • echo $::env(SYNTH_BUFFERING) :== 1
  • echo $::env(SYNTH_SIZING) :== 0
  • echo $::env(SYNTH_DRIVING_CELL) :== sky130_fd_sc_hd__inv_8

• Change the SYNTH_STRATEGY to optimize delay and enable cell sizing instead of buffering.

  • set ::env(SYNTH_STRATEGY) "DELAY 1"
  • set ::env(SYNTH_SIZING) 1

• Comparison of synthesis results with the modified settings

  Before	After
  Chip area for module '\picorv32a': 147950.646400 tns -3232.44 wns -26.53	Chip area for module '\picorv32a': 209179.369600 tns -266.36 wns -2.95

  The wns and tns values look much better and easier to fix later on.

23.2 Timing analysis with ideal clocks using openSTA

Note: We have already gone through STA basics previously. We will capture the important essentials once again here.

Setup timing analysis and introduction to flip-flop setup time

• Considering Ideal clocks, the condition to meet the setup timing for a single-cycle path (using same clock at launch and capture flops) is: T_clk > tc_q + t_comb + t_setup where,

  • T_clk : Ideal Clock Period
  • tc_q : Clock-to-Q Delay of the FF
  • t_comb : Total Combination Delay (including the wire/ net delay) from the output pin of the launch flop till the input pin of the capture flop.
  • t_setup : Setup time requirement of the FF

• Origin of FF setup-time

  • The minimum time required for the relevant internal nodes of the Flip-Flop circuit to capture the value at the "D" input pin, before the arrival of the clock triggering edge.
  • The exact values depend on the FF implementation style. If we consider a Mux/ Latch-based implementation of an edge-triggerted FF (where a negative latch is followed by a positive latch), the setup time is the time required for the 'D' input to reach the internal node, QM (i.e., the delay of the first Mux/ Latch)

• Clock Jitter

  • Clock jitter refers to the temporal variation of the clock period at a given point — that is, the clock period can reduce or expand on a cycle-by-cycle basis.
  • It is strictly a temporal uncertainty measure and is often specified at a given point on the chip.
  • Jitter can be measured and cited in one of many ways.
    • Cycle-to-cycle jitter refers to time varying deviation of a single clock period and for a given spatial location i is given as: Tjitter,i(n) = [(Ti, n+1 - Ti,n) - TCLK], where
      • Ti,n is the clock period for period n
      • Ti, n+1 is clock period for period n+1
      • and TCLK is the nominal clock period

• Setup Uncertainty

  • Clock uncertainty specifies a window within which a clock edge can occur.
  • With respect to setup timing, we specify a new parameter called setup uncertainty to model several factors like clock jitter, additional margins and clock skew (at pre-cts stage).

Lab: Configure OpenSTA for Post-synth timing analysis

• During the PnR flow, it is very much possible that some timing violations may get fixed, some violations will get better, some worse and new violations could also be introduced.

• In any PnR flow, the separate timing tool (like PrimeTime) is usually invoked outside of the automated flow for performing the timing analysis and timing ECO generation.

• In the OpenLANE flow, we use OpenSTA tool for Post-synthesis timing analysis.

• OpenSTA config file: pre_sta.conf:

  set_cmd_units -time ns -capacitance fF -current uA -voltage V -resistance kOhm -distance um
  
  read_liberty -max ./designs/picorv32a/src/sky130_fd_sc_hd__slow.lib
  read_liberty -min ./designs/picorv32a/src/sky130_fd_sc_hd__fast.lib
  
  #read_verilog ./designs/picorv32a/src/picorv32a.v
  read_verilog ./designs/picorv32a/runs/latest_21-03/results/synthesis/picorv32a.synthesis.v
  
  link_design picorv32a
  read_sdc ./designs/picorv32a/src/my_base.sdc
  
  check_setup -verbose
  
  report_checks -path_delay min_max -fields {nets cap slew trans input_pins}
  report_tns
  report_wns
  

• SDC file: my_base.sdc:

  set ::env(CLOCK_PORT) clk
  set ::env(CLOCK_PERIOD) 12.000
  set ::env(SYNTH_DRIVING_CELL) sky130_fd_sc_hd__inv_8
  set ::env(SYNTH_DRIVING_CELL_PIN) Y
  set ::env(SYNTH_CAP_LOAD) 17.65
  create_clock [get_ports $::env(CLOCK_PORT)]  -name $::env(CLOCK_PORT)  -period $::env(CLOCK_PERIOD)
  set IO_PCT  0.2
  set input_delay_value [expr $::env(CLOCK_PERIOD) * $IO_PCT]
  set output_delay_value [expr $::env(CLOCK_PERIOD) * $IO_PCT]
  puts "\[INFO\]: Setting output delay to: $output_delay_value"
  puts "\[INFO\]: Setting input delay to: $input_delay_value"
  
  
  set clk_indx [lsearch [all_inputs] [get_port $::env(CLOCK_PORT)]]
  #set rst_indx [lsearch [all_inputs] [get_port resetn]]
  set all_inputs_wo_clk [lreplace [all_inputs] $clk_indx $clk_indx]
  #set all_inputs_wo_clk_rst [lreplace $all_inputs_wo_clk $rst_indx $rst_indx]
  set all_inputs_wo_clk_rst $all_inputs_wo_clk
  
  
  # correct resetn
  set_input_delay $input_delay_value  -clock [get_clocks $::env(CLOCK_PORT)] $all_inputs_wo_clk_rst
  #set_input_delay 0.0 -clock [get_clocks $::env(CLOCK_PORT)] {resetn}
  set_output_delay $output_delay_value  -clock [get_clocks $::env(CLOCK_PORT)] [all_outputs]
  
  # TODO set this as parameter
  set_driving_cell -lib_cell $::env(SYNTH_DRIVING_CELL) -pin $::env(SYNTH_DRIVING_CELL_PIN) [all_inputs]
  set cap_load [expr $::env(SYNTH_CAP_LOAD) / 1000.0]
  puts "\[INFO\]: Setting load to: $cap_load"
  set_load  $cap_load [all_outputs]
  

• Invoke OpenSTA from another terminal and provide above config file as the input: sta pre_sta.conf

Lab: Optimize Synthesis to reduce setup violations

• In addition to the synthesis configuration variables that we have seen before, there are a few more that we can use to optimize synthesis to improve setup slack.
• If there are setup timing violations (and possible slew & max cap violations) from nets with high fanout, we can limit the fanout to improve hte delay using:

  set ::env(SYNTH_MAX_FANOUT) 4
  

• To view the nets driven by the output pin of a cell, the following command can be used:

  report_net -connections <net_name>
  

Lab: Steps to do basic Timing ECO

• From analysing the setup violations in OpenSTA, we will be able to infer the possible reasons for the violations
• One common reason could a large output slew for a net due to large capacitance load/ fanout which the synthesis tool could not optimize further. In this case, we can upsize the cell (i.e., replace the cell instance with a higher drive strength version of it) to reduce the delay using the replace_cell command.

  Syntax : replace_cell replace_cell instance lib_cell
  Example: replace_cell _44195_ sky130_fd_sc_hd__inv_4
           where, _44195_ is the instance name of the cell to be replaced
                  sky130_fd_sc_hd__inv_4 is the upsized std cell version
  

  • To check if the violation has been resolved:

    report_checks -from <instance or pin> -to <instance or pin> -through <instance> -path_delay max
    Example: report_checks -from <instance or pin> -to <instance or pin> -through _44195_ -path_delay min_max
    

• The above step of upsizing the cell to improve timing would obviously change the netlist, which needs to be updated in the netlist file for it to be captured in the OpenLANE flow. To write the updated netlist:

  write_verilog $OPENLANE_HOME/designs/picorv32a/runs/latest_21-03/results/synthesis/picorv32a.synthesis.v
  

• Note:
  • Fixing timing violations by ECOs is an iterative cyclical process.
  • The STA engineer(s) will do the necessary modifications like upsizing, replacing cell with a different Vt cell, inserting buffers etc. to fix a violation and provide the ECO to the PnR engineer(s).
  • The PnR engineer(s) will then take this ECO and do the PnR and perform the post-route timing analysis with the back-annotated values that capture the parasitics as well.
  • This may fix or improve the existing the timing violation, while there is a chance that it may introduce new violations.
  • Now the STA engineer(s) will take the new data and perform STA analysis again and provide new timing ECOs for the new violations.
  • This "spinning" process goes on till all voilations are rectified.

23.3 Clock Tree Synthesis using TritonCTS and Signal Integrity

Clock Tree Synthesis is the process of connecting the clocks to the clock pins of all sequential elements in the design by using inverters/ buffers in order to balance the skew and to minimize the insertion delay.

Ideal Clock Tree before CTS	Real Clock tree (H-tree) after CTS

Image courtesy: https://www.physicaldesign4u.com/2020/02/clock-tree-synthesis.html

• QoR parameters for a Clock Tree:
  1. Clock Insertion Delay/ Latency: Refers to the arrival time of the clock signal at the sink pin with respect to the clock source.
  2. Clock Skew: Refers to the difference in the clock arrival time between two sinks. It can further be sub-divided into Local Clock Skew and Global Clock Skew:
     • Local Clock Skew – The difference in the arrival times of the clock signal reaching any pair of registers that have a valid timing path between them.
     • Global Clock Skew – The difference in the arrival times of the clock signal reaching any pair of registers that may or may not have a valid timing path between them.
  3. Clock Slew (Transition Time): Clock slew needs to be as small as possible to provide a sharp timing edge by reducing the timing uncertainty. However this will have th implication of higher area and power if we use larger clock tree buffers.
  4. Duty Cycle: Unequal rise and fall times of the clock buffers is the primary cause of duty cycle distortion in a clock tree. Usually inverters are used instead of buffers to reduce DCD in a clock tree.
  5. Pulse Width: Usually SRAMs, flip-flops and latches will have minimum pulse width requirements to meet their internal timing. There will be minimum pulse width requirements for both the high and low times of a clock period.

Clock tree routing and buffering using H-Tree algorithm

• H-Tree algorithm

  • This is a clock tree routing algorithm that tries to minimize the skew by minimizing the routing length.
  • The clock routing taked place in the shape of the capital letter "H"
  • Algorithm Steps:
    1. Find out all the flops being clocked by the clock signal under consideration.
    2. Find out the center of all the flops.
    3. Trace clock port to center point.
    4. Now divide the core into two parts, trace both the parts and reach to each center.
    5. Then from this center, again divide the area into two and again trace till center at both the end.
    6. Repeat this algorithm till the time we reach the flop clock pin.
  • This is demonstrated in the following GIF file from wikipedia page:

• Clock buffering

  • To ensure the clock signal reaching each sink pin is having the required target slew/ transition time, we need to add clock repeaters or clock buffers at multiple points of the distribution network, ensuring the RC wire load is split across multiple levels of buffers.
  • To reduce Duty Cycle Distortion (DCD), clock buffers need to have equal rise and fall times. Usually it is very difficult to design buffers with equal rise and fall times and in a long clock tree with multiple levels of buffering, this can often lead to DCD. Hence instead of clock buffers, clock inverters are used in clock trees to reduce the introduction of DCD.

Clock Signal Integrity: Crosstalk and Clock Net Shielding

• Crosstalk Glitch

  • If a high slew net is somehow routed near to the clock net, a transition on this agressor net can cause a glitch on the the clock net while the clock net is at a logic LOW or HIGH level.
  • This happens due to the capacitive coupling between the nets and can cause the signal level on the clock net to temporarily go above the VIH or VIL level resulting in a spurious unwanted high/ low pulse on this clock net.
  • Depending on the height (and width) of the glitch, it could be a safe or unsafe one.
    • If the glitch height within the noise margin, it can be considered a safe glitch.
    • If the glitch height is above the noise margin limits, it is a potentially unsafe glitch.
    • If the glitch heigh is within the noise margin limits, it is an unpredictable case as the output transition may be random and could cause some flop to go metastable.
  • A glitch on the clock signal can have serious ramifications on the operation of the whole system and it can result in timing failure, writing of an invalid data to some flops or cause it to go metastable, eventually even resulting in system failure.
  • Theoretically, the reverse could also happen wherein the clock net can act as an aggressor since the edge rate of clock nets are usually very high.

• Crosstalk Delta Delay

  • Crosstalk delay occurs when both aggressor and victim nets switch together.
  • There can be two cases:
    1. When both the aggressor and victim nets transition in the same direction
       • The transition of the victim net becomes faster, being aided by the crosstalk, reducing its rise/ fall time and thus the cell delay.
    2. When the aggressor and victim nets transition in opposite direction
       • Transition of the victim net gets becomes delayed due to the crosstalk trying to oppose the transition on the victim net, causing a voltage bump on its rising or falling edge. This increases the transition time and hence the cell delay.
  • If the victim net is a clock net, crosstalk can cause a previously skew-balanced clock tree to become unbalanced. Since the clock skew is affected, it may result in some paths getting faster or slower resulting in hold or setup violations respectively.

• Clock Net Shielding

  • The critical clock nets are shielded to prevent any unwanted crosstalk on to the clock nets.
  • Shielding nets are placed on either sides of the clock net along its entire length, to decouple the aggressor net. The shielding nets are connected to either VDD or GND (either both of them to VDD (or GND), or one of them to VDD & the other to GND). (Basically the shielding nets need to be connected to a non-transitioning net, low impedance upon which an aggressor has no effect).

Lab: Steps to run CTS using TritonCTS

• Command to run cts: run_cts
• After CTS, a new netlist <design_name>.synthesis_cts.v will be created in the runs/<tag>/results/synthesis/ folder that includes the information on the generated clock clock tree and the newly instanced clock buffers.

Lab: Steps to verify CTS runs

• CTS configuration variables to verify:

Configuration Variable	Details
CTS_TARGET_SKEW	The target clock skew in picoseconds, usually 10% of the clock period.
CTS_ROOT_BUFFER	The name of cell inserted at the root of the clock tree (sky130_fd_sc_hd__clkbuf_16 in our case)
CTS_TOLERANCE	An integer value that represents a tradeoff of QoR and runtime. Higher values will produce smaller runtime but worse QoR.
LIB_CTS	The liberty file used for CTS. By default, this is the LIB_SYNTH_COMPLETE minus the cells with drc errors.
CTS_MAX_CAP	Defines the maximum capacitance for clock tree synthesis in the design in pF.

23.4 Timing Analysis with real clocks using OpenSTA

Setup timing analysis using real clocks

Hold timing analysis using real clocks

Lab: Steps to analyze timing with real clocks (Post-CTS STA) using OpenSTA

• In OpenRoad, the timing analysis is performed by creating a db file using the LEF and DEF files of the design.
• db creation is a one-time process (unless the def changes). To create the db, invoke OpenRoad from within the OpenLANE shell using openroad. And then from within the OpenRoad shell execute the following commands:

  read_lef /openLANE_flow/designs/picorv32a/runs/latest_21-03/tmp/merged.lef
  read_def /openLANE_flow/designs/picorv32a/runs/latest_21-03/results/cts/picorv32a.cts.def
  write_db picorv32a_cts.db
  

• Performing STA:

  read_db picorv32a_cts.db
  read_verilog /openLANE_flow/designs/picorv32a/runs/latest_21-03/results/synthesis/picorv32a.synthesis_cts.v
  read_liberty $::env(LIB_SYNTH_COMPLETE)
  ##read_liberty -max $::env(LIB_SLOWEST)
  ##read_liberty -min $::env(LIB_FASTEST)
  link_design picorv32a
  read_sdc /openLANE_flow/designs/picorv32a/src/my_base.sdc
  set_propagated_clock [all_clocks]
  report_checks -path_delay min_max -format full_clock_expanded -digits 4 -fields {net cap slew input_pins fanout}
  

• Be sure to perform the timing analysis with the correct library file which was used for CTS (which was the LIB_SYNTH_COMPLETE or the LIB_TYPICAL in our case).
• Note: As of now, CTS does not support multi-corner optimization.

Lab: Steps to observe impact of bigger CTS buffers on setup and hold timing

• Modify the CTS_CLK_BUFFER_LIST variable to exclude the sky130_fd_sc_hd__clkbuf_1 cell and re-run CTS again.
• Be sure to modify the CURRENT_DEF variable to point to the DEF file after placement before triggering the CTS run.

  % echo $::env(CTS_CLK_BUFFER_LIST)
  sky130_fd_sc_hd__clkbuf_1 sky130_fd_sc_hd__clkbuf_2 sky130_fd_sc_hd__clkbuf_4 sky130_fd_sc_hd__clkbuf_8
  
  set ::env(CTS_CLK_BUFFER_LIST) [lreplace $::env(CTS_CLK_BUFFER_LIST) 0 0]
  set ::env(CURRENT_DEF) /openLANE_flow/designs/picorv32a/runs/latest_21-03/results/placement/picorv32a.placement.def
  
  run_cts
  

• We will be able to see the setup and hold slacks having some amount of improvement, but do note that this comes with a potentially large area & power penalty due to the larger clock buffers used.

---

Day 24: Final steps for RTL2GDS using tritonRoute and openSTA

24.1 Routing and Design Rule Check (DRC)

Introduction to Maze Routing - Lee's Algorithm

• Lee's algorithm is one possible solution for maze routing problems based on breadth-first search.
  • If a path exists between the source and target, this algorithm guarantees finding it.
  • It always finds the shortest path between the source and target.
• Steps
  1. The layout area is divided into a Routing grid with keep-off areas for the pre-placed macros and logical blockages for the I/O pad ring.
  2. The next step is to identify the source pin (S) and target pin (T) of the cells in the grid
  3. Now, based on the distance from the grid box "S", adjacent grid boxes (only horizontal and vertical grid boxes are considered adjacent and not the diagonal ones) are labelled as 1. Next, the grid boxes adjacent to those labelled 1 are now labelled as 2. This iterative process progresses until we hit the target pin's, "T" grid box.
  • Note: The wave expansion marks only points in the routable area of the chip, not in the blocks or already wired parts.
• Routes with lower number of bends are preferred.

Lee's Routing Algorithm The blue route is preferred over the black one due to lower number of bends

• This algorithm, however, has a high cost in term of both memory usage and run time.
• To overcome these short-comings, there are other more advanced algorithms like Line Search algorithm & Steiner tree algorithm.

Design Rule Check during routing

• While performing routing, the router tool needs to follow the DRC rules related to routing provided by the PDK.
• Basically, the origin of these DRCs come from the manufacturing process used like photolithography or others, that for example, define the minimum wire width that can be manufactured, the minimum spacing between two wire in a metal layer etc.

Typical DRC rules for wires like width, spacing and pitch

• Check out the DRC rules related to the local interconnect and the metal layers for sky130 PDK DRC.

A Signal short during route Fixing it using other metal layers gives rise to new DRC checks related to VIAs

24.2 Power Distribution Network and Routing

Lab: Steps to build Power Distribution Network

• Levels of Power distribution

  > VDD, VSS pins/ balls
    |--> VDD, VSS pads
         |--> Core Power Rings
              |--> VDD, VSS Horizontal, Vertical Straps
                   |--> Standard Cell Rails
                   |
                   |--> Macro Power Ring
  

Source: Power Distribution Network

• Command to generate PDN in openLANE: run_power_grid_generation

Layout after PDN generation
Layout zoomed

Lab: Steps to perform routing

• To run the routing in OpenLANE, execute: run_routing

No DRC violations after routing
No Timing violations at TYPICAL corner
Layout after routing (KLayout)
Layout after routing - Zoomed (Magic)

• Post-route STA

  read_lef /openLANE_flow/designs/picorv32a/runs/latest_25-03/tmp/merged.lef
  read_def /openLANE_flow/designs/picorv32a/runs/latest_25-03/results/routing/picorv32a.def
  read_spef /openLANE_flow/designs/picorv32a/runs/latest_25-03/results/routing/picorv32a.spef
  
  write_db picorv32a_routing.db
  
  read_db picorv32a_routing.db
  read_verilog /openLANE_flow/designs/picorv32a/runs/latest_25-03/results/synthesis/picorv32a.synthesis_preroute.v
  read_liberty $::env(LIB_SYNTH_COMPLETE)
  ##read_liberty -max $::env(LIB_SLOWEST)
  ##read_liberty -min $::env(LIB_FASTEST)
  link_design picorv32a
  read_sdc /openLANE_flow/designs/picorv32a/src/my_base.sdc
  set_propagated_clock [all_clocks]
  report_checks -path_delay min_max -format full_clock_expanded -digits 4 -fields {net cap slew input_pins fanout}
  

Basics of global and detail routing

• Global route using fast route and Detailed route using tritonRoute
• Global route provides a routing guide.
• In Detailed route, algorithms are used to find the best possible connectivity using the routing guides from global route.

TritonRoute Features

• Performs initial detail route
• Honors the preprocessed route guides (obtained after global route) - i.e., attempts as much as possible to route within route guides.
• Assumes route guides for each net satisfy inter-guide connectivity.
• Works on proposed MILP-based panel routing scheme with intra-layer parallel and inter-layer sequential routing framework.

---

Day 25 - Post-placement STA analysis of your Design (RISC-V Myth CPU Core)

25.1 OpenLANE Installation (latest stable version)

For ease of installation, OpenLane uses Docker images, that include all the required applications, binaries and the flow scripts.
The OpenLane Installation Documentation page explains in detail the step to setup OpenLANE. A short summary of the necessary steps is given below:

• Install the required dependencies:

  sudo apt-get update
  sudo apt-get upgrade
  sudo apt install -y build-essential python3 python3-venv make git
  

• Docker Installation
  1. Run the following command to uninstall all conflicting packages:

     for pkg in docker.io docker-doc docker-compose docker-compose-v2 podman-docker containerd runc; do sudo apt-get remove $pkg; done
     

  2. Set up Docker's apt repository

     # Add Docker's official GPG key:
     sudo apt-get update
     sudo apt-get install ca-certificates curl
     sudo install -m 0755 -d /etc/apt/keyrings
     sudo curl -fsSL https://download.docker.com/linux/ubuntu/gpg -o /etc/apt/keyrings/docker.asc
     sudo chmod a+r /etc/apt/keyrings/docker.asc
     
     # Add the repository to Apt sources:
     echo \
       "deb [arch=$(dpkg --print-architecture) signed-by=/etc/apt/keyrings/docker.asc] https://download.docker.com/linux/ubuntu \
       $(. /etc/os-release && echo "$VERSION_CODENAME") stable" | \
       sudo tee /etc/apt/sources.list.d/docker.list > /dev/null
     sudo apt-get update
     

  3. Install the Docker packages:

     sudo apt-get install docker-ce docker-ce-cli containerd.io docker-buildx-plugin docker-compose-plugin
     

  4. Verify that the Docker Engine installation is successful by running:

     sudo docker run hello-world
     

     To make Docker available without root:

     sudo groupadd docker
     sudo usermod -aG docker $USER
     sudo reboot # REBOOT!
     
     # After reboot
     docker run hello-world
     

• Download OpenLane from GitHub:

  git clone --depth 1 https://github.com/The-OpenROAD-Project/OpenLane.git
  cd OpenLane/
  make
  make test
  

  • These steps will download and build OpenLane and sky130 PDK.
  • Finally, it will run a ~5 minute test that verifies that the flow and the PDK were properly installed. Successful test will output the following line:

  Basic test passed
  

25.2 Basic Usage

25.2.1 Invoking the OpenLane Docker session:

cd to the OpenLane folder

# Start the Docker session:
make mount

# Leave the Docker
exit

25.2.2 Adding Your Designs

• To add a new design, the following command creates a configuration file for your design:

  # JSON Configuration File
  ./flow.tcl -design <design_name> -init_design_config -add_to_designs
  
  # Tcl Configuration File
  ./flow.tcl -design <design_name> -init_design_config -add_to_designs -config_file config.tcl
  

  This will create the following directory structure:

  designs/<design_name>
  ├── config.json (or config.tcl)
  

  • NOTE: The <design_name> must match the top-level module name of your design exactly. Otherwise, OpenLane will throw an error (at least by the run_synthesis stage).
• It is recommended to place the verilog files of the design in a src directory inside the folder of the design as following:

  designs/<design_name>
  ├── config.tcl
  ├── src
  │   ├── design.v
  

25.2.3 Running the flow for the design:

• To run the automated flow:

  ./flow.tcl -design <design_name>
  

• To run the flow interactively (Interactive Mode):

  ./flow.tcl -interactive
  

  A tcl shell will be opened where the openlane package is to be sourced

  package require openlane 0.9
  

  Now, the following main commands run the various major steps of the Physical design flow:

  0. Any valid Tcl code
  1. prep -design <design> [-tag TAG] [-config CONFIG] [-init_design_config] [-overwrite]
  2. run_synthesis
  3.run_floorplan
  4.run_placement
  5.run_cts
  6.run_routing
  7.a write_powered_verilog 
  7.b set_netlist $::env(routing_logs)/$::env(DESIGN_NAME).powered.v
  8. run_magic
  9. run_magic_spice_export
  10. run_magic_drc
  11. run_lvs
  12. run_antenna_check
  

25.2.4 Some FYI Notes:

• Command-Line Arguments that can be passed to flow.tcl

Argument	Description
-tag <name> (Optional)	Specifies a “name” for a specific run. If the tag is not specified, a timestamp is generated for identification of that run.
-overwrite (Optional)	Flag to overwrite an existing run with the same tag
-synth_explore	If enabled, synthesis exploration will be run (only synthesis exploration), which will try out the available synthesis strategies against the input design.
-verbose <level> (Optional)	0: Outputs only high-level messages 1: Also ouputs some of the inner workings of the flow scripts >=2: Forwards outputs from all tools

• Design configuration files: config.json (or config.tcl)
  Each OpenLane-compatible design must come with a configuration file written in either JSON or Tcl.
• Flow Configuration variables
• PDK Configuration variables
• Tcl commands available in OpenLane
  • When running the flow in interactive mode, use the -verbose <level> flag with the prep -design <design_name> command to set the verbose output level (2 and greater outputs everything including tool outputs). Example: prep -design riscv_core -tag run1 -overwrite -verbose 2

25.3 Synthesis Results

• Some synthesis settings modified:

  set ::env(SYNTH_STRATEGY) "DELAY 0"
  set ::env(SYNTH_BUFFERING) 1
  set ::env(SYNTH_SIZING) 1
  set ::env(SYNTH_SPLITNETS) 0
  set ::env(SYNTH_MAX_FANOUT) 6
  

• In addition, the Openlane/scripts/yosys/synth.tcl file was modified, with the sequence of yosys/ abc synthesis commands being modified in accordance with this comment from Yosys issue #4266.

• Synthesis cell usage statistics:

  === riscv_core ===
  
     Number of wires:               8458
     Number of wire bits:          11999
     Number of public wires:         176
     Number of public wire bits:    3717
     Number of memories:               0
     Number of memory bits:            0
     Number of processes:              0
     Number of cells:              11997
       sky130_fd_sc_hd__a2111o_2      10
       sky130_fd_sc_hd__a2111oi_2      2
       sky130_fd_sc_hd__a211o_2      132
       sky130_fd_sc_hd__a211oi_2      38
       sky130_fd_sc_hd__a21bo_2        3
       sky130_fd_sc_hd__a21boi_2       6
       sky130_fd_sc_hd__a21o_2        41
       sky130_fd_sc_hd__a21oi_2      894
       sky130_fd_sc_hd__a221o_2      442
       sky130_fd_sc_hd__a221oi_2       3
       sky130_fd_sc_hd__a22o_2       318
       sky130_fd_sc_hd__a22oi_2        5
       sky130_fd_sc_hd__a2bb2o_2       4
       sky130_fd_sc_hd__a311o_2       54
       sky130_fd_sc_hd__a311oi_2       2
       sky130_fd_sc_hd__a31o_2        72
       sky130_fd_sc_hd__a31oi_2       41
       sky130_fd_sc_hd__a32o_2       216
       sky130_fd_sc_hd__a32oi_2        2
       sky130_fd_sc_hd__a41o_2         6
       sky130_fd_sc_hd__a41oi_2        1
       sky130_fd_sc_hd__and2_2        40
       sky130_fd_sc_hd__and2b_2       24
       sky130_fd_sc_hd__and3_2       104
       sky130_fd_sc_hd__and3b_2        7
       sky130_fd_sc_hd__and4_2        32
       sky130_fd_sc_hd__and4b_2       23
       sky130_fd_sc_hd__and4bb_2      28
       sky130_fd_sc_hd__buf_1       2242
       sky130_fd_sc_hd__buf_2        940
       sky130_fd_sc_hd__conb_1       775
       sky130_fd_sc_hd__dfxtp_2     1907
       sky130_fd_sc_hd__inv_2         80
       sky130_fd_sc_hd__mux2_2        78
       sky130_fd_sc_hd__nand2_2      105
       sky130_fd_sc_hd__nand3_2       17
       sky130_fd_sc_hd__nand3b_2       2
       sky130_fd_sc_hd__nand4_2        1
       sky130_fd_sc_hd__nand4b_2       1
       sky130_fd_sc_hd__nor2_2       118
       sky130_fd_sc_hd__nor3_2         1
       sky130_fd_sc_hd__nor3b_2        5
       sky130_fd_sc_hd__nor4_2        12
       sky130_fd_sc_hd__nor4b_2        5
       sky130_fd_sc_hd__o2111ai_2      3
       sky130_fd_sc_hd__o211a_2       77
       sky130_fd_sc_hd__o211ai_2      31
       sky130_fd_sc_hd__o21a_2       944
       sky130_fd_sc_hd__o21ai_2       74
       sky130_fd_sc_hd__o21ba_2      516
       sky130_fd_sc_hd__o21bai_2      32
       sky130_fd_sc_hd__o221a_2       31
       sky130_fd_sc_hd__o221ai_2       4
       sky130_fd_sc_hd__o22a_2         5
       sky130_fd_sc_hd__o22ai_2        2
       sky130_fd_sc_hd__o2bb2a_2      10
       sky130_fd_sc_hd__o2bb2ai_2      4
       sky130_fd_sc_hd__o311a_2       19
       sky130_fd_sc_hd__o311ai_2       1
       sky130_fd_sc_hd__o31a_2       935
       sky130_fd_sc_hd__o31ai_2        1
       sky130_fd_sc_hd__o32a_2        12
       sky130_fd_sc_hd__o41a_2        45
       sky130_fd_sc_hd__or2_2         75
       sky130_fd_sc_hd__or2b_2        37
       sky130_fd_sc_hd__or3_2         63
       sky130_fd_sc_hd__or3b_2        18
       sky130_fd_sc_hd__or4_2        133
       sky130_fd_sc_hd__or4b_2        22
       sky130_fd_sc_hd__or4bb_2        4
       sky130_fd_sc_hd__xnor2_2       27
       sky130_fd_sc_hd__xor2_2        33
  

• Chip area for module '\riscv_core': 115293.075200

25.4 Placement (Pre-CTS)

• Placement optimizations:

  Reading design constraints file at '/openlane/designs/riscv_core/src/riscv_base_post_cts.sdc'…
  [INFO]: Setting output delay to: 4.0
  [INFO]: Setting input delay to: 4.0
  [INFO]: Setting load to: 0.01765
  [INFO]: Setting clock uncertainty to: 0.25
  [INFO]: Setting clock transition to: 0.15
  [INFO]: Setting timing derate to: 5.0 %
  [INFO]: Setting RC values...
  [INFO RSZ-0027] Inserted 1 input buffers.
  [INFO RSZ-0028] Inserted 10 output buffers.
  [INFO RSZ-0058] Using max wire length 8182um.
  [INFO RSZ-0034] Found 80 slew violations.
  [INFO RSZ-0035] Found 14 fanout violations.
  [INFO RSZ-0036] Found 11 capacitance violations.
  [INFO RSZ-0038] Inserted 303 buffers in 94 nets.
  [INFO RSZ-0039] Resized 9488 instances.
  [INFO RSZ-0042] Inserted 513 tie sky130_fd_sc_hd__conb_1 instances.
  [INFO RSZ-0042] Inserted 262 tie sky130_fd_sc_hd__conb_1 instances.
  

• Design area: 124358 u^2 33% utilization

25.5 STA Comparison: Post-synthesis vs. Post-Placement (Pre-CTS)

Post-Synthesis	Post-Placement (Pre-CTS)

---

Day 26 - Post-CTS, Post-Routing STA analysis of your Design

26.1 STA Comparison

OpenLane Configuration:

• config.tcl
• sky130A_sky130_fd_sc_hd_config.tcl
• Pre-CTS SDC File
• Post-CTS SDC File

Design Area

• Synthesis: 115293.075200
• Placement: 124358 u^2 33% utilization
• CTS: 129960 u^2 34% utilization

Post-Synthesis	Post-Placement (Pre-CTS)
Post-Routing	Post-CTS

NOTE: Writing powered verilog netlist

write_powered_verilog -output_def riscv_core_powered_verilog.def -output_nl $::env(routing_logs)/$::env(DESIGN_NAME).nl.v -output_pnl $::env(routing_logs)/$::env(DESIGN_NAME).powered.v
set_netlist $::env(routing_logs)/$::env(DESIGN_NAME).powered.v

---