I build report-backed RTL and verification projects with SystemVerilog, UVM, Verilator, PyTorch, C/C++/SystemC reference models, assertions, formal checks, Python automation, and open-source implementation tools. The portfolio is deliberately separated by verification scope rather than repeating the same claims across projects.
| Project | Primary specialty | Selected measured evidence | Review |
|---|---|---|---|
| RISC-V Chiplet SoC | Firmware-driven subsystem integration, DMA, UPF, low power, CDC | 70 / 70 stable; 85 / 85 GCC/ISS; 26 / 26 power; 4 / 4 CDC | Metrics · CI · v2.0.1 |
| AXI4 L1 Cache DV | Cache microarchitecture, C++ replay, SECDED RAS, bounded MSI, SRAM BIST | 22 / 22 directed; 127 / 127 replay; 25 / 25 MSI seeds; 2 / 2 integrated BIST | Metrics · CI · v0.4.0 |
| AXI4 QoS Fabric DV | Reusable UVM/VIP, AXI concurrency, SystemC replay, QoS/fairness | 8 / 8 UVM; 130 / 130 replay; 24 / 24 advanced crosses; 72 / 72 QoS points | Metrics · CI · v0.3.1 |
| INT8 Tensor Accelerator DV | PyTorch-to-RTL checking, quantization, AXI integration, RV32I offload | 130 / 130 exact comparisons; 40 / 40 benchmark; 25 / 25 correctness; 5 / 5 mutations | Metrics · CI · v0.3.1 |
| Area | Chiplet SoC | L1 cache | QoS fabric | INT8 accelerator |
|---|---|---|---|---|
| System integration | RV32 firmware, APB MMIO, DMA, AES service | CPU/cache/AXI memory path plus bounded two-cache MSI | Four-initiator/four-target shared fabric | GCC RV32I, APB/AXI control, and streaming INT8 offload |
| Verification methodology | Procedural closure plus supporting UVM/RAL | C++ trace replay and mutation-driven debug | Principal real-UVM lane and reusable AXI agents | Exact integer oracle and tagged scoreboard |
| Architecture depth | UPF, retention/isolation, async CDC | Replacement, maintenance, associativity, SECDED, SRAM BIST | IDs, out-of-order responses, QoS, aging, fairness | Signed MACs, requantization, ReLU, saturation |
| Independent models | Python/C transaction and CRC models | Cycle-independent C++ cache model | SystemC/TLM arbitration and routing model | PyTorch tensor and nn.Linear models |
| Evidence | Firmware, power, formal, CDC, coverage | Stress, C++ replay, RAS, MSI mutations, BIST | UVM, VIP self-test, QoS dashboard, CDC, formal | PyTorch differential checks, RV32 cycle benchmark, mutations, latency |
flowchart LR
SOC["RISC-V chiplet\nfirmware + low power"] --> PORT["Hardware DV portfolio"]
L1["AXI4 L1 cache\nmicroarchitecture + RAS"] --> PORT
FAB["AXI4 QoS fabric\nUVM/VIP + concurrency"] --> PORT
ML["INT8 tensor accelerator\nPyTorch + numerical DV"] --> PORT
All headline metrics are generated from checked-in canonical reports. These projects demonstrate open-source engineering evidence, not UCIe/AXI certification or commercial UPF, CDC, timing, and formal signoff.