About Etched Etched is building the world’s first AI inference system purpose-built for transformers - delivering over 10x higher performance and dramatically lower cost and latency than a B200. With Etched ASICs, you can build products that would be impossible with GPUs, like real-time video generation models and extremely deep & parallel chain-of-thought reasoning agents. Backed by hundreds of millions from top-tier investors and staffed by leading engineers, Etched is redefining the infrastructure layer for the fastest growing industry in history.

Key Responsibilities

• Kernel-Mode Driver Development: Design, develop, and maintain kernel-mode drivers ensuring high reliability, informative debug, and optimal performance.
• Performance Optimization: Analyze and optimize driver performance for demanding AI workloads, focusing on minimizing latency and maximizing throughput.
• Hardware Integration and Co-Design: Collaborate closely with hardware engineers throughout the ASIC design process..
• Virtualization Support: Implement driver support for device virtualization technologies, including SR-IOV, VFIO, and para-virtualization.
• Memory Management: Implement efficient memory management strategies considering kernel memory mapping, page tables configuration, NUMA awareness for device data caching, and IOMMU configuration.
• Security: Build kernel drivers fundamentally designed to support and maintain security across host processes, physical memory spaces, and device attestation.
• Debugging and Troubleshooting: Diagnose and resolve complex driver-related issues, using common kernel debugging tools and techniques (ftrace, dmesg, etc.) to identify and fix bugs.
• Synchronization and Concurrency: Design and implement synchronization mechanisms to handle concurrent access to multiple accelerators.
• System Validation and Testing: Develop and execute comprehensive test plans to validate driver functionality, stability, and performance in manufacturing and in general production environments.
• Collaboration and Troubleshooting: Collaborate with software and hardware teams to diagnose and resolve complex system-level issues.

Representative Projects

• Develop and optimize kernel-mode drivers for new ML accelerators.
• Implement and optimize memory management, including kernel memory mapping and IOMMU configurations, for high-bandwidth data transfers.

Must-Have Skills And Experience

• Proficiency in C/C++.
• Strong understanding of kernel-mode driver development and debugging.

Nice-to-Have Skills And Experience

• Experience with GPU driver development.
• Experience with CUDA, OpenCL, or other GPU programming models.
• Experience with performance profiling and benchmarking tools (perf, VTune).
• Knowledge of hardware virtualization techniques, including para-virtualization.
• Experience with CI/CD pipelines.
• Experience with Rust.
• Experience with ML frameworks like Tensorflow or Pytorch.
• Experience with data center orchestration technologies (Kubernetes, Docker).

Ideal Background

• Candidates with experience in developing and debugging kernel-mode drivers for GPU or other accelerator devices.
• Candidates with a strong understanding of hardware/software interactions.
• Candidates with experience in optimizing driver performance for demanding workloads.
• Candidates with experience in ML workloads.
• Candidates who have debugged complex hardware and software interactions, especially in virtualized environments.
• Candidates with experience in implementing and optimizing SR-IOV and VFIO.
• Candidates with in-depth knowledge of kernel memory mapping, page tables, and IOMMU.
• Candidates with experience in hardware-software co-design projects.

Benefits

• Full medical, dental, and vision packages, with generous premium coverage
• Housing subsidy of $2,000/month for those living within walking distance of the office
• Daily lunch and dinner in our office
• Relocation support for those moving to West San Jose

Compensation Range

• $150,000 - $275,000

How We’re Different Etched believes in the Bitter Lesson. We think most of the progress in the AI field has come from using more FLOPs to train and run models, and the best way to get more FLOPs is to build model-specific hardware. Larger and larger training runs encourage companies to consolidate around fewer model architectures, which creates a market for single-model ASICs.

We are a fully in-person team in West San Jose, and greatly value engineering skills. We do not have boundaries between engineering and research, and we expect all of our technical staff to contribute to both as needed.