Principal Architect, Performance Analysis and Modeling

The role

The role involves analyzing and modeling performance across the hardware/software boundary of our AI inference accelerators. This includes emerging hardware technologies like DIMC, D2D, and 3D-DRAM, and emerging workloads such as generative inference and multi-modal Large Language Models (LLMs).

You will work closely with Hardware Design, Compiler, Inference Server, and Kernel teams, translating workload analysis into concrete modeling inputs and surfacing HW/SW improvement opportunities.

What You Will Do

• Analyze emerging ML workloads, multi-modal LLMs, CoT reasoning models, and video/audio generation to identify performance-relevant properties.
• Build and maintain analytical performance models that project behavior on current and future d-Matrix hardware generations.
• Develop and extend architecture simulators to support performance analysis of proposed HW/SW features.
• Partner with hardware design, compiler, inference server, kernel, and product teams to validate modeling assumptions and surface downstream implications.
• Track relevant ML architecture and algorithms research and incorporate findings into modeling work.
• Propose targeted HW/SW feature improvements based on modeling results and workload analysis.
• Document modeling methodology and findings for reuse across the architecture team.

What You Will Bring

• BSEE with 6+ years of industry experience, or MSEE with 4+ years of industry experience.
• Working knowledge of computer architecture, HW/SW co-design, performance modeling, and ML fundamentals (particularly DNNs).
• Programming fluency in C/C++ or Python.
• Experience building or working with analytical performance models or architecture simulators.
• Self-motivated and collaborative, comfortable working across hardware and software teams.

Preferred Qualifications

• Experience optimizing AI/ML workloads on accelerator technologies.
• Research and investigation in AI/ML architecture/microarchitecture.

Pay

Commensurate with experience.

Schedule

3 days per week hybrid. Will consider remote in the United States.