Skip to content
View 3more102's full-sized avatar

Block or report 3more102

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
3more102/README.md

I design, implement, and verify digital hardware systems, with a focus on processor architectures, SoC integration, hardware accelerators, and verification infrastructure. My work also extends into machine learning from first principles, connecting algorithm-level understanding with hardware-efficient computation.

Important

Core Engineering Principle: A design is not validated simply because it passes its tests. Its correctness must be supported by independent, reproducible evidence. Across my projects, I use reference-model co-simulation, formal verification, mutation testing, protocol checking, and post-synthesis regression to challenge hardware implementations and assess the strength of their verification environments.

// Verification Sign-Off Invariant
property p_justified_confidence;
    @(posedge clk)
    passing_tests |-> (independent_oracle_match && formal_proven && mutation_qualified);
endproperty

assert property (p_justified_confidence)
    else $error("Verification Gap: Passing tests alone do not prove correctness.");

⚑ Engineering Architecture

System and Verification Architecture Flowchart

πŸ” Expand Technical Focus Matrix (7 Core Engineering Domains)
Domain Areas of Work
Digital RTL Design RISC-V processors, pipelined datapaths, accelerators, memory interfaces
Design Verification Golden-model comparison, constrained testing, scoreboards, mutation testing
Formal Methods Model checking, protocol properties, exhaustive state-space exploration
SoC Architecture Processor–accelerator integration, AXI interconnects, cache coherence
ASIC Implementation RTL-to-GDSII workflows, SKY130HD, PPA evaluation, DFT
FPGA Prototyping Functional validation and implementation on physical FPGA platforms
Hardware-Oriented ML Quantized inference, systolic computation, neural networks from scratch

πŸ› οΈ Featured Engineering Projects

Processor Architecture & SoC Integration

πŸ”Ή Cascade SoC

Cascade SoC Status 4x Verification

RISC-V SoC INT8 Accelerator SKY130HD DFT

RISC-V System-on-Chip with an integrated INT8 accelerator, bringing together processor architecture, accelerator functionality, and system-level verification.

  • Integration verified through four independent approaches
  • Design-for-test (DFT) implementation with measured fault coverage
  • RTL-to-GDSII implementation targeting the SKY130HD standard-cell library
  • Status: Physical implementation and place-and-route in progress
View System Topology
β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”      β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚  RISC-V Core  │◀────▢│ System Interconnect  β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜      β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
                                  β”‚
                       β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β–Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
                       β”‚   INT8 Accelerator   β”‚
                       β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
                                  β–Ό
 [4 Independent Verification Approaches] + [DFT & SKY130HD PnR]

πŸ”Ή Cascade RV32I

FPGA Booted Verification Scope

RV32IM 5-Stage Pipeline FPGA Booted Mutation Tested

Five-stage RV32IM processor verified across RTL simulation, synthesized logic, and physical FPGA execution.

  • Five-stage pipelined RISC-V microarchitecture
  • Architectural commit-log co-simulation against an independent golden model
  • Gate-level regression to assess synthesized behavior
  • Mutation testing to evaluate verification effectiveness
  • Successfully booted on physical FPGA hardware
View Cross-Level Validation Flow
[RV32IM RTL] ──▢ Commit-Log Co-Sim ◀── [Independent Golden Model]
     β”‚
     β”œβ”€β”€β–Ά Mutation Testing (Testbench Strength Evaluation)
     β”œβ”€β”€β–Ά Synthesis ──▢ Gate-Level Regression
     └──▢ Physical FPGA Bitstream ──▢ Hardware Boot Verified

Hardware Acceleration & ASIC Design

πŸ”Ή Systolic Tile

64 MACs per cycle Bit-Exact

8Γ—8 INT8 64 MACs/cycle Cyclone V Zynq-7010

8Γ—8 INT8 systolic-array accelerator built for deterministic arithmetic correctness and portable hardware implementation.

  • 64 multiply–accumulate operations per cycle
  • Bit-exact validation against a Python reference model
  • FPGA builds targeting Intel Cyclone V and AMD/Xilinx Zynq-7010

SKY130HD PPA Bit-Exact

INT8 Inference SKY130HD GDSII PPA Measured

Hardware implementation of a quantized LeNet-5 neural-network inference datapath connecting quantized ML computation with measurable ASIC results.

  • Bit-exact comparison against an independent Python oracle
  • Physical implementation and measured power, performance, and area using SKY130HD

Formal Verification & Protocol Correctness

πŸ”Ή Coherex

Model Checked SC Oracle

4-Core MESI Model Checking Sequential Consistency

Four-core MESI cache-coherence system focused on coherence invariants, memory-ordering correctness, and independent checker validation.

  • RTL implementation of a multicore coherence system
  • Exhaustive protocol model checking
  • Independent sequential-consistency oracle
  • Mutation-qualified checkers for verification-quality assessment
View Coherence Verification Topology
 [Core 0 L1]   [Core 1 L1]   [Core 2 L1]   [Core 3 L1]
      β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
                    [MESI Coherence Fabric]
                           β”‚
        β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
        β–Ό                  β–Ό                  β–Ό
 [Model Checker]   [SC Memory Oracle]  [Mutation Checkers]

Unbounded Proof 27/27 and 14/14

AXI4 / AXI4-Lite 2Γ—2 Crossbar Unbounded Formal

AXI4 / AXI4-Lite 2Γ—2 crossbar verified through simulation, mutation testing, and formal proof.

  • 27 / 27 regression tests passing
  • 14 / 14 injected mutation defects detected
  • Unbounded formal proofs achieved for both interconnects

Mutation-Driven Verification & First-Principles ML

36/38 Mutants 2 Survivors

IΒ²C Master RTL Mutation Testing 36/38 Detected

IΒ²C master RTL and mutation-driven verification environment measuring what a testbench can detect rather than relying on passing-test counts alone.

  • 38 deliberately injected RTL defects
  • 36 defects detected by the verification environment
  • Two surviving mutations explicitly demonstrated and documented

MNIST Accuracy No Autograd

First Principles Pure NumPy No Autograd

Neural-network implementation using NumPy without automatic differentiation, built for mathematical understanding, numerical correctness, and reproducible experimentation.

  • 99.39% MNIST classification accuracy (99.55% with test-time augmentation)
  • Manually implemented backpropagation, convolution, and batch normalization
  • No autograd framework

πŸ“Š Selected Measured Results

Project Sign-Off Evidence Verified Result
Cascade SoC 4x Verif Four independent integration-verification approaches; DFT fault coverage; SKY130HD PnR active
Cascade RV32I FPGA Boot Golden-model and gate-level verification; mutation tested; physical FPGA boot
Systolic Tile 64 MACs 64 INT8 MAC operations/cycle; bit-exact reference-model comparison
LeNet-5 Accelerator SKY130HD Bit-exact validation; SKY130HD PPA measurements
Coherex Model Checked Exhaustive protocol model checking and mutation-qualified checkers
AXI Interconnect Unbounded Proof 27/27 tests passing; 14/14 mutation defects detected; unbounded formal proofs
MutantHunt IΒ²C Mutation Score 36/38 injected defects detected; two survivors documented
MNIST from Scratch Accuracy 99.39% accuracy; 99.55% with test-time augmentation

Each result is specific to its project's documented implementation, configuration, and verification scope.


πŸ”¬ Verification Philosophy

I treat verification as an engineering discipline with measurable outcomes, not simply a final stage of development. The objective is not to maximize the number of passing tests. It is to maximize justified confidence in the design.

Five-Stage Verification Methodology Flowchart

  • 01 β€” Independent Reference Models: Compare DUT behavior against separate software or architectural models to avoid relying exclusively on implementation-derived expectations.
  • 02 β€” Formal Reasoning: Use formal properties and exhaustive model checking where applicable to investigate behavior beyond directed simulation scenarios.
  • 03 β€” Verification of the Verification Environment: Inject deliberate defects and determine whether existing checkers detect them. Surviving mutations are evidence of verification limitations that require explanation.
  • 04 β€” Cross-Level Validation: Where supported by the project, validate behavior across RTL simulation, synthesis, gate-level regression, and physical FPGA implementation.
  • 05 β€” Measurable Engineering Evidence: Report actual test results, proof status, performance measurements, implementation results, and known limitations.

πŸš€ Current Development

Milestone Target Platform Pipeline Stage Objective
Cascade SoC β€” Silicon Implementation SKY130HD PnR Advancing the integrated RISC-V SoC through physical implementation to complete the RTL-to-GDSII flow on SKY130HD
Keccak Tile β€” Cryptographic Hardware Tiny Tapeout RTL Progressing from a validated golden model toward synthesizable RTL, independent verification, and a physical implementation submission

These are ongoing engineering milestones, not completed silicon claims.


πŸ“ˆ Engineering Telemetry & Interests

  • RISC-V microarchitecture and SoC design
  • Digital ASIC design and RTL verification
  • Formal verification and verification-quality measurement
  • Memory systems, interconnects, and cache coherence
  • Hardware acceleration for machine learning and cryptography
  • DFT, physical implementation, and silicon validation

Note

Build the design. Build an independent way to challenge it. Measure the evidence. Document the limits. The repositories above reflect an effort to develop hardware systems that are not only functional, but also independently verifiable, reproducible, and supported by concrete engineering results.

Footer Wave Banner

Pinned Loading

  1. cascade-rv32i cascade-rv32i Public

    A 5-stage pipelined RV32I core in Verilog, verified by golden-model commit-log co-simulation, gate-level regression and mutation testing.

    Verilog 2

  2. cascade-soc cascade-soc Public

    RISC-V SoC (cascade-rv32i) + INT8 systolic-array accelerator (systolic-tile), integration-verified and DFT'd on sky130hd; RTL-to-GDSII backend flow in progress

    Verilog

  3. Coherex Coherex Public

    Cache coherence verification: exhaustive MESI protocol model checking, a 4-core snoopy RTL cluster, litmus tests judged against an exhaustive sequential-consistency oracle, and mutation-qualified c…

    Python

  4. lenet5-asic-accelerator lenet5-asic-accelerator Public

    Verified LeNet-5 int8 inference datapath in SystemVerilog, with a bit-exact Python oracle and real sky130hd PPA (pre-layout).

    SystemVerilog 2

  5. mnist-nn-scratch mnist-nn-scratch Public

    A neural network built from NumPy and nothing else - hand-written backprop, convolutions, batch norm and optimizers. 99.39% on MNIST, 99.55% with test-time augmentation.

    Python

  6. systolic-tile systolic-tile Public

    8x8 INT8 weight-stationary systolic array - 64 MAC/cycle, verified bit-exact against a Python golden model, with FPGA self-test builds for Cyclone V and Zynq-7010.

    SystemVerilog