Software Engineer - Hardware Abstraction Layer, AWS Machine Learning Accelerators

Custom silicon chips live at the heart of AWS machine learning servers (Trainium and Inferentia), and enable machine learning (ML) for AWS's customers. We’re looking for engineers to scale the system software team developing the hardware abstraction layer (HAL) that manages these cutting-edge ML system-on-chips (SoCs). The HAL forms the lowest level of the AWS infrastructure management software stack. As a software engineer developing the SoC HAL, you will: - Work with hardware designers to build HALs for newly developed SoC IPs - Work with system software teams to solve SoC and system-level architectural issues, drive debug, architect the HAL itself, and innovate on cross-functional solutions - Continuously test and deploy your software stack to multiple internal customers - Refactor and maintain existing codebases throughout the device lifecycle - Innovate on the tooling you provide to customers, making it easier for them to use our SoCs AWS's Annapurna Labs organization designs and deploys some of the largest custom silicon in the world, with many subsystems that must all be managed, tested, and monitored. The SoC HAL is a critical piece of the AWS infrastructure management software stack that ensure the chip is functional, performant, and secure. You will thrive in this role if you: - Are strong in C++ and familiar with Python - Enjoy working with hardware-based systems, and diving into chip and system architecture - Know how to build effective abstractions over low-level SoC details - Have strong opinions about software architecture, and are able to apply them effectively - Are familiar with modular driver architectures (such as the Linux or Windows device-driver stacks) - Enjoy learning new technologies, building software at scale, moving fast, and working closely with colleagues as part of a small, startup-like team within a large organization Although we build and deploy ML chips, no ML background is needed for this role. You (and your software) won’t be doing ML. Our HAL lives at the lowest level of the backend AWS infrastructure responsible for managing our ML servers. You and your team will develop HALs for components used by machine learning, like PCIe and HBM, but won’t need to deeply understand ML yourselves. This role can be based in either Cupertino, CA or Austin, TX. The team is split between the two sites, with no preference for one over the other. This is a fast-paced role where you'll work with thought-leaders in multiple technology areas. You'll have high standards for yourself and everyone you work with, and you'll be constantly looking for ways to improve your software, as well as our products' overall performance, quality, and cost. We're changing an industry. We're searching for individuals who are ready for this challenge, who want to reach beyond what is possible today. Come join us and build the future of machine learning! A day in the life A few videos help explain what the Annapurna Labs ML team is working on: - https://youtu.be/4nfkonjjICo?si=nKhM1Wv4108mnIOa - https://youtu.be/n38WDflRbjQ?si=SZEV5i_5du1jKYP- About the team Annapurna Labs, designs custom silicon powering AWS’s cloud infrastructure. Custom SoCs live at the heart of Amazon ML servers — including Inferentia and Trainium Systems — delivering high-performance ML inference and training at cloud scale.