This is a cryptography accelerator ASIC designed using the SKY130 (130nm) process node, to be taped out on the Google/Skywater/efabless open-source MPW shuttle. It includes hardware implementations of cores for AES128/AES256 and SHA256, as well as an experimental VGA-based game demo (it also includes the "caravel" debug-harness/padframe which is being used in this tapeout). The eventual goal of this project is to use it for embedded/IoT security applications.
This repository (asinghani/crypto-accelerator-builds) contains the openlane-based build environment (including the final tapeout-ready GDS files). It contains the main project repository (asinghani/crypto-accelerator-chip) as a submodule and uses symbolic links to form the directory structure (this is done to keep the main project repository cleaner, as this repository is quite unwieldy to manipulate due to its size). The build products must be stored inside this repo in order to use it with the efabless Open MPW platform, which pulls the build products directly from the git repository.
When working with the design, this repository must always be used as it contains fundamental parts of the build environment.
The accelerator itself (which is designed to be usable on both FPGA and ASIC) is designed in Chisel3 and is at asinghani/crypto-accelerator (it is also submoduled into this repository to make builds easier).
This chip contains 4 major components:
-
AES128/256 Accelerator
- Optimized for balance between speed and area, can encrypt or decrypt a 16-byte block in ~20 cycles for AES128 or ~28 cycles for AES256
- Supports both ECB and CBC modes (CBC is recommended because it is much more secure against certain types of attacks) with no performance penalty for using CBC
-
SHA256 Accelerator
- Can hash a single 512-bit block in ~66 cycles (and is able to process multiple blocks immediately back-to-back, i.e. when hashing a large file)
-
VGA Game Demo (experimental)
- Original design was contributed by Ethan Polcyn (and then ported to run on the ASIC environment)
- Consists of a small infinite side-scrolling jumping game (based on the Chromium Browser's "No internet"-game), playable using button input from one of the chip's I/O pins
- Renders VGA video output at 640x480 through the chip's output pins
- Caravel harness/padframe
- Required as part of Open MPW tapeout - contains a picorv32 CPU (which is used to interface with the rest of the cores through the Wishbone bus) and the padframe
- See efabless/caravel for details.
- Python 3
- Icarus Verilog (for simulations)
- OpenLANE rc6
- Magic VLSI Layout Tool (>= 8.3.60)
- (Optional) Scala & SBT (only needed to regenerate the full build from the Chisel3 sources - otherwise can just use the prebuilt
crypto_accelerator.vfile that is already in the repository)
# Clone repo (and submodules)
git clone --recurse-submodules https://github.com/asinghani/crypto-accelerator-builds
# Uncompress the gzipped build products and harness files
make uncompressTo run the tests (both RTL-level and gate-level):
cd verilog/dv/caravel/accelerator/
# To run a specific test case:
cd <aes / dino_vga / sha256>
make SIM=RTL SPOOF_FAST_FLASH=1 # for accelerated RTL test
make SIM=RTL # for normal RTL test
make SIM=GL # for gate-level test
# To run all AES and SHA test cases (from inside `accelerator` directory):
make rtl # RTL tests
make gl # gate-level testsThe aes and sha256 testbenches are self-checking, and will print failure messages if there are any issues. There are additional verification steps in the Chisel3 testbenches (in the crypto-accelerator repo), including several hundred more test cases.
The dino_vga test is special because it cannot be automatically verified. Instead, there is a python script in the dino_vga directory (named parse.py, with numpy, vcdvcd, and opencv-python modules as dependencies) which can be invoked (from inside the dino_vga directory) as python3 parse.py dino_vga.vcd "dino_vga_tb.dump[40:0]" to parse the VCD and generate a PNG file frame.png containing the VGA frame outputted by the design (which can be visually inspected for correctness).
(Expects that the download procedure has been completed, including the uncompress stage)
# Optional (only if need to re-generate crypto_accelerator.v from Chisel3 sources)
cd crypto-accelerator-chip && ./update-verilog.sh && cd ..
# Build top-level accelerator
cd openlane && make accelerator_top && cd ..
ls openlane/accelerator_top/runs/ # Find the most recent run name
# Move the netlists & build files
./migrate-accel-gds.sh <path to run including openlane/accelerator_top/runs/>
./migrate-gl.sh <path to run including openlane/accelerator_top/runs/>
# Build the "user project wrapper" which fits the accelerator into the caravel padframe
cd openlane && make user_project_wrapper && cd ..
# Re-run the netlist migration since it includes user_project_wrapper as well
./migrate-gl.sh <path to run including openlane/accelerator_top/runs/>
# Add user_project_wrapper into caravel padframe
make ship
# At this point, the top-level GDS (caravel + accelerator) is generated in `gds/caravel.gds`
# Optionally, to render the caravel to an image using KLayout
./render.sh # May need to run with XVFB if in a headless environmentThe relevant files in this repository are as follows (the caravel harness files are excluded here, see details about them in efabless/caravel):
add_spdx.py - Script to add SPDX license identifiers to source files
compress.py - Compress large build products using gzip to fit into GitHub repo size limit
crypto-accelerator-chip - Submodule which contains the main source code (no build products)
└─crypto-accelerator - Submodule which contains the Chisel3 design for the crypto accelerator
def - Post-routing DEF build files (for the accelerator as well as the caravel harness)
gds - Final GDS files (for the accelerator as well as the caravel harness)
info.yaml - Metadata used by efabless platform for tapeout
lef - Abstracted LEF views (for the accelerator as well as the caravel harness)
mag - Magic build files (for the accelerator as well as the caravel harness)
maglef - Abstracted magic build files (for the accelerator as well as the caravel harness)
Makefile - Script to build the final "ship" product for caravel
migrate-accel-gds.sh - Script to transfer the GDS files after running an openlane build
migrate-dv.sh - Script to re-build the directory structure in `verilog/dv/caravel/accelerator`
migrate-gl.sh - Script to transfer the gate-level netlists to be used for gate-level simulations
ngspice - Simulations of the analog blocks in the caravel harness
openlane
├─accelerator_top - Configuration for building the main accelerator block
└─user_project_wrapper - Configuration for building the "wrapper" block which is used to fit the accelerator into the harness
qflow - Analog components from the caravel harness
render.py - KLayout script to generate a top-level render of caravel.gds
render.sh - Shell script to invoke KLayout and load render.py to generate the render
scripts - Various scripts used to prepare the chip for tapeout
spi - LVS extraction files
utils - Various scripts to handle file format conversions
verilog
├─dv
│ └─caravel
│ └─accelerator
│ ├─aes - Test cases for the AES128/256 core
│ ├─dino_vga - Simulation for the VGA game demo (must be verified visually)
│ ├─Makefile - Runs all simulations
│ └─sha256 - Test cases for the SHA256 core
│
├─rtl
│ ├─accelerator
│ │ ├─accelerator_top.v - Top-level design, instantiates and connects all the relevant modules
│ │ ├─crypto_accelerator.v - Crypto accelerator verilog (auto-generated from Chisel3 RTL design)
│ │ └─dino - Directory which contains all the sources for the VGA game demo
│ │
│ │
│ └─user_project_wrapper.v - Instantiation of the accelerator core (exposing relevant ports to caravel)
│
├─gl - Autogenerated gate-level models of each of the modules
└─stubs - Stubs for LVS on analog components of caravel harness