Machine Learning SoC Architect

Meta•Austin, TX

2d•$212,000 - $294,000

About The Position

Meta is seeking a Machine Learning SoC Architect for its Silicon Engineering organization responsible for building custom silicon solutions that power the infrastructure underpinning Meta's AI and data center workloads at scale. As an ASIC Engineer specializing in architecture, performance and modeling, you will define and drive the architectural definition, performance analysis, pre-silicon modeling, and microarchitectural exploration of custom ASICs designed for Meta's Data Centers. In this role, you will own ASIC architecture specification, establish the performance modeling methodology and long-term silicon roadmap strategy, partnering with other silicon, and software teams to ensure Meta's infrastructure silicon meets the demanding throughput, latency, and efficiency targets required at hyperscale.

Requirements

Bachelor's degree in Computer Science, Computer Engineering, relevant technical field, or equivalent practical experience
Experience and knowledge of Computer Architecture concepts such as microprocessor architecture, memory systems, on-chip interconnection networks, hardware/software partitioning etc
12+ years of prior experience in defining and delivering multiple high performance ASICs into production, with focus on architecture definition and performance analysis
Experience in ASIC performance modeling, microarchitectural analysis, or pre-silicon simulation for custom silicon or SoC designs
Proficiency in C++ and Python for developing simulation models, automation frameworks, and performance analysis tools
Experience with performance analysis of data center, AI accelerator, or high-performance computing workloads on custom silicon
Experience defining architecture and microarchitectural specifications and driving cross-functional alignment across architecture, RTL, and physical design teams

Nice To Haves

Familiarity with post-silicon performance validation and model-to-hardware correlation methodologies
Programming in C or C++ with knowledge of mapping hardware algorithms to efficient C/C++ code
Master's or PhD degree in Electrical Engineering, Computer Engineering or related field
Domain knowledge in one or more of power/performance tradeoffs, ML networks, ML frameworks such as Pytorch
Experience building or scaling performance modeling infrastructure for hyperscale data center ASICs, including network, storage, or AI inference accelerator designs

Responsibilities

Work on algorithm analysis, performance analysis and architecture definition of Machine Learning ASICs
Map Data Center workloads to heterogeneous ASICs that contain multiple different programmable processors and hardware accelerators. Perform detailed calculations to specify computation throughput, memory bandwidth and latency; evaluate performance v/s area v/s power tradeoffs
Drive the architecture definition of one or more of the following ASIC sub-systems: compute, memory, Network-On-Chip (NoC), collectives, debug etc. and chiplet based multi-die SoCs
Identify appropriate workloads and micro-benchmarks to be used for performance analysis and drive this analysis on simulation and emulation platforms to define and validate the architecture
Evangelize your innovative architectural solutions with your peers and leadership, while mentoring members of the architecture team
Collaborate with cross functional teams working on RTL design, Design Verification, Firmware/Software development, Pre-Post silicon validation and Program Management to deliver first pass functional silicon on an aggressive schedule
Collaborate with software and firmware teams to ensure that the ASIC meets end to end application performance goals while maintaining ease and efficiency of software development