Allreduce IP Verification
- Led verification of Tesla’s first-generation custom Allreduce engine from IP through multi-chip emulation, achieving 100% functional coverage with waivers.
- Architected a multi-platform Python stimulus generator and golden reference model capable of producing valid multi-chip stimulus and checkers for both simulation and emulation.
- Developed a scalable UVM testbench from scratch with reusable checkers, scoreboards, and coverage models.
- Owned test planning, regression triage, and closure of all coverage metrics
- Partnered with RTL and performance modeling teams to detect and resolve key performance bottlenecks ahead of schedule.
DMA Subsystem Verification
- Co-developed the DMA subsystem testbench ensuring correct inter-block synchronization across all test modes.
- Built end-to-end checkers enabling early detection of functional bugs, significantly reducing debug cycles.
- Architected a scalable multi-DMA verification environment spanning multiple SoCs, exposing critical cross-SOC corner-case issues.
RandGen – Standalone SOC Stimulus Generator
- Designed a Python-based random stimulus framework leveraging architectural specifications to produce stimulus and memory vectors, routing data, and checker collateral for DMA verification.
- Created reusable libraries for memory management, preload utilities, system routing, and instruction randomization — greatly reducing test development time.
Slice IP Verification
- Led verification of a custom memory fabric interconnect linking multiple internal NOCs to the memory controller.
- Built a UVM environment with drivers, monitors, checkers, and constrained-random stimulus to achieve full functional and code coverage.
Emulation
- Developed self-checking multi-SOC test flows for Cadence Palladium, reducing manual debug effort.
- Created synthesizable modules to trigger hard-to-reach corner-case scenarios.