N
Neurophos Tech Stack
Photonic AI chips delivering exascale inference at GPU power budgets
Neurophos builds optical processors for AI inference, compressing the compute density of large server racks into a single GPU-sized package. The company is hiring 27 engineers—mostly at lead and principal levels—across a tight stack of silicon photonics, ASIC design, and compiler work (PyTorch, JAX, Triton). Current projects center on GEMM optimization, 4-bit quantization, and porting large language models to optical inference, suggesting a near-term push toward production LLM acceleration.
Tech Stack 30 technologies
Core StackJenkins GitLab CI/CD GitHub Actions Python Linux PyTorch GitHub Hugging Face C++ Cadence Virtuoso ASIC SerDes FPGA Triton JAX SystemC UEFI Secure Boot Open Compute Project Verilator SystemVerilog Silicon Photonics C UVM Xcelium JasperGold Git Cadence Xilinx Vivado PCIe
What Neurophos Is Building
◆Challenges
- Power wall
- Rising energy consumption and costs
- Power and scalability challenges
- Outpacing energy capacity
- Energy capacity constraints
- High energy consumption in data centers
- Energy capacity bottleneck
- Physical limits of silicon
- Scalability challenges in data centers
- Energy capacity shortfall
▲Active Projects
- Optical gemm engine modeling
- Developing quantization strategies for 4-bit precision
- Custom processor core design for opu
- Optimizing gemm operations
- Llm accelerator compiler development
- Porting llms to optical inference engines
- Trace-driven simulation infrastructure
- Performance modeling infrastructure
- Sram vector processor modeling
- System-on-chip programs
Hiring Activity
30 roles · 20 in 30d
Department
Seniority
Notable leadership hires: Director of Marketing, Functional Modeling Lead, Tech Lead, Optics Director
Company intelligence
