Remote job
Computer Arithmetic Engineer, Datapath Design
Job details
About this role
Role overview Design and evaluate arithmetic execution units for a new processor, including decisions about how to meet workload needs within area, latency, and energy constraints. The work spans datapath design, quantitative cost modeling, implementation analysis, and close collaboration with research and pipeline engineering to turn promising ideas into practical hardware.
Responsibilities - Design arithmetic datapaths such as adders, multipliers, dividers, floating-point units, and mixed-precision blocks. - Compare implementations using ROM, SRAM, logic, or recomputation, and quantify their area and energy tradeoffs. - Build and refine cost models so design decisions can be checked against evidence. - Work with researchers on number representations and error behavior, translating results into hardware and providing feedback on feasible directions. - Run synthesis, timing, and power analysis, then iterate based on tool results. - Coordinate interfaces and latency requirements with pipeline designers, including changes to the surrounding machine where useful.
Requirements - Production experience designing arithmetic hardware that shipped, beyond coursework or early research prototypes. - Strong knowledge of IEEE 754, including nuanced behavior and edge cases. - Ability to reason quantitatively about area, timing, and power tradeoffs. - Ability to explain technical concepts clearly to engineers outside arithmetic design.
Benefits and work setup The role is remote within the United States, with several hours of daily overlap with Pacific time. Remote employees should expect about one week per month on site during the first year; travel and standard accommodation are covered. Relocation support may be available. The position is exempt and includes paid time off, access to available health plans and retirement benefits, and a possible equity opportunity during negotiation. Early work includes evaluating an existing block with a cost model and recommending whether its current form makes sense.