commit | a0c4b42dec831761c03e03f8fe028eb6ffa38a5c | [log] [tgz] |
---|---|---|
author | Harrison Pham <harrison@harrisonpham.com> | Tue Nov 09 19:27:37 2021 -0800 |
committer | Harrison Pham <harrison@harrisonpham.com> | Tue Nov 09 19:27:37 2021 -0800 |
tree | 1bc802ea8248606595880c86028f0457a84f65a5 | |
parent | 717abf6cf68134273cceb65487a0cd3ef89d015f [diff] |
Slow down to 50 MHz
Randsack is a test chip for trying out random number generators and PUFs.
digitalcore_macro
- Digital top sea of gates containing control logic and digital peripherals.gpio0
- Wishbone 32-bit GPIO peripheralpwm[0-3]
- Wishbone PWM peripherals with 16-bit prescaler and 16-bit counter/compareuart0
- Wishbone UART peripheralring0
- Ring oscillator controller for collapsing ring.ring1
- Ring oscillator controller for free running ring oscillator.collapsering_macro
- Trimmable collapsing ring oscillators for generating random numbers with a configurable output divider. See ip/randsack/sch/collapsering.sch
xschem schematic for design.ringosc_macro
- Trimmable ring oscillator.All custom IP blocks are in located in the ip/randsack/
directory. Third party IP is in the ip/third_party/
directory.
The more analog-like blocks like the ring oscillators are designed using stdcells in xschem and simulated with ngspice.
Due to limited time all blocks are synthesized using the standard openlane flow instead of hand layout. The resulting netlist is inspected to ensure minimal modifications by the tools. The resulting extracted spice file is then simulated.
Unfortunately the process to do backannotated timing sims (SDF) doesn‘t appear simple. Hope is the small macros are small and that any delays are tiny and don’t cause issues.
All simulations are performed at tt/ff/ss corners to ensure reasonable performance across PVT.
A bunch of knobs are built into the design to minimize risk. All blocks feature many trim bits and output dividers in case performance ends up being too fast for the synthesized digital control blocks.
The output of the oscillator blocks can be muxed to output GPIOs for debug. GPIOs are limited to ~60 MHz so the internal clock dividers should be used.
Use the mpw-3
tag in the https://github.com/efabless/OpenLane.git repo. As of this time the Docker Hub repo is missing the mpw-3
tag so manually set the openlane tag to master
which currently points to the same commit. See envsetup
for required environment vars.
All the *_macro
blocks need to be hardened first before finally hardening the user_project_wrapper
macro.
See the verilog/dv/randsack*
directories for RTL/GL testbenches.