tree: 5be62766538641d55f1d861a755d664d00e463d0 [path history] [tgz]
  1. .travisCI/
  2. def/
  3. docs/
  4. gds/
  5. lef/
  6. macros/
  7. mag/
  8. maglef/
  9. ngspice/
  10. openlane/
  11. qflow/
  12. scripts/
  13. signoff/
  14. spi/
  15. utils/
  16. verilog/
  17. .travis.yml
  18. build
  19. info.yaml
  20. LICENSE
  21. Makefile
  22. Makefile.master
  23. manifest
  24. mpw-one-b.md
  25. README.md
  26. ship.log
  27. user_proj_example.log
  28. user_project_wrapper.log
README.md

SKY130 SHA3 Miner Caravel SOC

miner

Table of Contents

Introduction

In mining a proof-of-work (POW) is used to verify the authenticity of a blockchain entry. What is a POW? A POW is a mathematical puzzle which is difficul to solve but easy to verify.

For this example a SHA3 mining core is defined for a hypothetical blockchain that uses the SHA3-256 hash. Finding a hash that meets certain conditions is difficult, verifying it does is simple.

The core is implemented on Skywater's SKY130 process curtesy of the Open MPW Shuttle Program sponsored by Google.

We are given:

  • H: 256 bit header (Fixed value input)
  • N: 64 bit nonce (The value we must find)
  • D: 256 bit difficulty (Problem difficulty, smaller value = more difficult)

The problem we need to solve is to find any value of N, such that the SHA3-256 hash of the nonce concatenated to the header, is less than or equal to the difficulty:

D >= SHA3({H, N})

NOTE: This will not mine a real blockchain. It intended as an exaple of hashing algorithm optimized for mining using a multi-stage permutation pipeline.

Implementation

This ASIC is generated using end-to-end open source EDA tools. A 12 stage pipeline design is used in two phases since a fully unrolled 24 stage pipeline exceeds the capacity of this ASIC. Each stage consists on an identical complex combinatorial chain of 1600 inputs, 6 control inputs and 1600 outputs. Each stage must render its 1600 output values, based on the inputs and altered according to the control inputs, within one clock cycle. Using 12 such linked stages and an appropriate feedback path from the last to the first we can generate a single hash per clock cycle one half of the time. The maximum clock speed will be determined by the propagation delay of a single stage. The number of SHA3 hashes per second (hash rate in mining parlance) that can be generated using this type of folded pipeline is given as follows:

F is the clock frequency in hertz S the number of stages (must be a divisor of 24. i.e., 1, 2 4 6 12, 24) H Hash rate

H = (F * S) / 24

A Wishbone client register file is implemented and serves for control and status by the Caravel picorv32 CPU core. This circuitry is conveniently clocked by the Wishbone bus clock. The 12 combinatorial stages however are clocked from a separate user programmable DLL clock, allowing hash rate adjustments. Proper synchronization is applied where timing domain crossing occurs.

In mining we do not really care what the winning hash is, we only care that it meets the difficulty requirement and what nonce was used to achieve it. The nonce is a continuously incrementing counter so we simply freeze it when a match is found.

At a high level the chip is intended to function as a low level controller for the SHA3 pipeline, communicating via the Caravel I2C or SPI ports to a larger computer to handle higher level functions such as Internet mining protocols.

Miner Component

The component is an Wishbone bus device with a 23 word memory mapped register file for control and status. All user project Verilog source is contained in the verilog/rtl/user_proj_example.v file.

User block interface.

SignalWidthDescription
wb_clk_i150MHz Wishbone bus clock (1-bit input)
wb_rst_i1Asynchronous reset (1-bit input)
wbs_stb_i1Select
wbs_cyc_i1Active bus cycle
wbs_we_i1Write enable
wbs_sel_i4Byte lane select
wbs_dat_i32Input data
wbs_adr_i32Address
wbs_ack_o1Bus ccycle acknowledge
wbs_dat_o32Output data
io_inMPRJ_IO_PADSIO pin bus input
io_outMPRJ_IO_PADSIO pin bus output
io_oebMPRJ_IO_PADSIO pin bus output enable
user_clock21300 MHz miner core clock

Register file

Each register occupies 4 bytes, starting at base address 0x30000000

Reg. #NameRead/WriteDescription
0-1SOLN_REGRO64-bit Solution
2STATUS_REGROStatus (see below)
3SHA3_REGROFingerprint “SHA3”
4-11HDR_REGRW256-bit Header
12-19DIFF_REGRW256-bit difficulty
20-21START_REGRW64-bit start nonce
22CTL_REGRWControl (see below)

Status register

Bit #NameDescription
0FOUNDSolution found. Solution is stored and status updated
1RUNNINGThe run ctl bit is set and the solution nonce is auto-incrementing
2TESTINGThe test ctl bit is set and compare diff equal (for verification only)

Control register

Bit #NameDescription
0RUN0 - clear, 1 - auto increment the solution nonce and check hashes
1TEST0 - normal mode, 1 - test mode, look for exact match with diff
2HALT0 - normal mode, 1 - halt mining and update status
23-16PAD_LASTlast pad byte, 0x80 for KECCACK-256 and SHA3-256
31-24PAD_FIRSTfirst pad byte, 0x01 for KECCACK-256, and 0x06 for SHA3-256

Verilog Module Hierarchy

Top module:  \user_proj_example
Used module:     \sha3_256_miner_core_12
Used module:         \sha3_256_miner_round
Used module:             \permutation
Used module:     \sha3_256_miner_regs

Building

Prerequisites

Caravel

Openlane

Magic

Google Skywater PDK

Synthesizing

From the command line:

git clone https://github.com/miscellaneousbits/caravel_sha3_256_crypto_miner.git
cd caravel_sha3_256_miner/openlane
make user_proj_example
make user_project_wrapper
cd ..
make ship

This will create the preliminary Skywater fab input artifacts. Essentially a giant GDS file containing a full physical description of the system on a chip, in this case over 300,000 logic cells and 1,000,000 copper traces.

NOTE: The entire process takes about 10 hours on a high end PC with plenty of memory.

Future

Currently this 12 stage pipeline design is fully autorouted and uses over 300,000 cells and is the most that can be crammed into the available die space. The layout is very sparse in order to get successful routing. It may be possible to optimize and harden smaller repeating blocks the place then manually to achieve a fully unrolled 24 stage pipeline. This would achieve twice the performance of the current version.

Picorv32 Firmware

TBD

Long live open-everything.