Devansh Patel Senior ASIC Design Engineer
Santa Clara, CA • devansh.patel@gmail.com • +1 408-555-0192
Profile Summary
- Senior ASIC Design Engineer with 10 years in front-end ASIC design for data center GPUs and 5G baseband chips, owning blocks from microarchitecture spec to A0 silicon on TSMC 4N and TSMC N7.
- Works every day in HDLs (SystemVerilog, Verilog), verification (UVM, Synopsys VCS), synthesis and timing (Design Compiler, PrimeTime), static checks (SpyGlass CDC, UPF) and buses and IP (AMBA AXI4, PCIe Gen5).
- Strongest in pipelined datapath design, timing closure above 1.5 GHz, low-power design and multi-clock CDC, with a spec-first design flow and coverage-driven sign-off that deliver RTL that hits its freeze date.
- Works hand in hand with architecture, DV, DFT and physical design teams on a tape-out schedule with fixed RTL freeze dates, from spec reviews and timing ECOs to post-silicon debug in the lab.
- Brought 4 new-grad engineers up to speed on lint-clean RTL and SDC writing, leads the cluster’s RTL code reviews, and owns the microarchitecture spec template the team still uses.
Technical Skills
- Languages & Scripting:
- SystemVerilog, Verilog, SystemC, Python, Tcl, Perl
- Simulation & Formal:
- UVM, Synopsys VCS, Synopsys Verdi, SystemVerilog Assertions, JasperGold, Cadence Xcelium
- Synthesis & STA:
- Design Compiler, PrimeTime, Fusion Compiler, SDC constraints, MCMM analysis, timing ECOs
- Lint, CDC & Power Intent:
- SpyGlass CDC, SpyGlass RDC, SpyGlass Lint, UPF, VC LP, Conformal LEC
- DFT & Test:
- Siemens Tessent, scan insertion, ATPG, MBIST, JTAG (IEEE 1149.1), IEEE 1687
- Buses & Interface IP:
- AMBA AXI4, PCIe Gen5, CHI, APB, HBM3, NVLink
- Process & Physical Design:
- TSMC 4N, TSMC N7, Cadence Innovus, Synopsys IC Compiler II, Siemens Calibre, StarRC
Education
Work Experience
- Wrote the microarchitecture spec and SystemVerilog RTL for the L2 cache request arbiter of a data center GPU on TSMC 4N, rebuilding it as a 6-stage pipeline with credit-based flow control that cut request latency from 9 to 6 cycles.
- Caught 41 RTL bugs ahead of freeze with a UVM environment for the arbiter that paired constrained-random traffic with SVA checkers and JasperGold proofs on the credit counters, run nightly on Synopsys VCS.
- Synthesized the arbiter in Design Compiler from SDC constraints written against the spec, then closed a -138 ps setup gap at 1.8 GHz through RTL retiming and signed off clean in PrimeTime across all 24 PVT corners.
- Wrote the UPF power intent for the L2 slice’s 3 power domains, adding fine-grained clock gating and power gating with retention flops, checked in VC LP, which cut idle power by 31% against the previous generation.
- Mapped every clock and reset crossing across 5 asynchronous domains in the memory subsystem, using async FIFOs, pulse synchronizers and a staged reset tree, and brought open SpyGlass CDC waivers down from 57 to 3 by sign-off.
- Integrated the licensed PCIe Gen5 controller and PHY into the GPU’s host interface, writing the AMBA AXI4 bridge and the IP-XACT register map, and had the integration regression clean 5 weeks ahead of the subsystem milestone.
- Partnered with DFT on scan insertion and MBIST for 48 SRAM macros in the L2 slice, rewriting the RTL that blocked scan access, and lifted Tessent ATPG stuck-at coverage from 96.1% to 99.3%.
- Handed timing budgets and floorplan-aware partitions to physical design for tape-out, then led A0 bring-up of the memory subsystem with JTAG scan dumps and on-chip trace, root-causing 7 silicon bugs and fixing them with metal-only ECOs in time for the A1 production stepping.
- Designed the Verilog RTL for the DMA descriptor engine of a 5G baseband processor on TSMC N7, moving packet buffers between LPDDR4 and the DSP clusters at 92% of peak bandwidth.
- Worked with physical design on floorplan-aware partitioning of the modem subsystem and closed 46 post-route timing ECOs, so the chip taped out in 15 months and reached production with no base-layer respin.
- Wrote the specs and register maps for 4 IPs that firmware and DV worked from, and onboarded 3 new-grad engineers to the team’s lint and code review flow.