Director of AI Hardware Abstraction

Flex Logix develops industry-leading AI inference engines and is the #1 provider of eFPGA solutions. Our InferX X1 chipset is the industry’s fastest and most-efficient AI edge inference accelerator. Our architecture is covered by dozens of patents and is highly differentiated, giving us a sustainable competitive advantage.

One significant advantage of the Flex solution is its synergistic combination of software and hardware, and a key part of that synergy is the Hardware Abstraction team– the interface between those two components. Flex Logix is searching for a Director of Hardware Abstraction reporting to the VP of Software. Currently the group is composed of five engineers, so the position is at present a hands-on managerial position. It is anticipated that the team will grow rapidly in the near future (one requirement for the successful candidate is the ability to ramp up engineering teams).

As Director of Hardware Abstraction, you will be responsible for driving the design, planning, and development of the Hardware Abstraction interface in the Flex Inference Compiler, which includes budgeting, staffing, and managing the team The Hardware Abstraction team responsibilities include:

• Determining methods of scheduling highly mathematical ML kernels (such as convolution and transformers) on InferX X1 hardware that achieve high performance within resource constraints.
• Developing and deploying Verilog that effects those schedules on the X1 hardware.
• Developing, extending, and maintaining a software interface that provides information about the abstraction to higher software levels, and that makes selections from that abstraction based on specified requests.
• Developing Verilog to initialize, configure, and reconfigure the InferX hardware.

Responsibilities:

While this is expected to grow into a broader managerial position, it is initially a technical management position, and as such will require technical understanding and hands-on work. More generally, the responsibilities include

• Leading a multi-site team (Mountain View, Austin, and Vancouver) that develops high quality highly-optimized ML kernels for the InferX X1. The team is primarily located in Mountain View for the moment, but there is no requirement that that continue in the future. As such, some travel will be required.
• Working with the Inference Compiler team to understand their needs with respect to kernels and information they need in order to select optimal performing kernels.
• Working with Marketing, Sales, and Software to develop custom product requirements and features, particularly with respect to key kernels and their performance requirements.
• Working with the Hardware Team to understand the key features of hardware, how to extract the best possible hardware performance, and to propose new features and changes to increase performance and simplify software development.
• Instituting processes for product development that produce high-quality, efficient kernels with minimal development effort and time.
• Mentor junior engineers, particularly in the aspects of developing high performance mathematical SW.

Required Abilities and Experience

• Bachelors or higher degree in computer science, electrical engineering, or a related field.
• 5+ years managing development of complex software products.
• Strong people skills, particularly in dealing with software and hardware engineers
• Demonstrated ability to hire and motivate bright, energetic software and hardware engineers.
• Ability to think strategically and tactically and to build consensus to make programs successful
• Experience in machine learning frameworks and high-performance computing.
• Excellent communications and technical presentation skills.
• An understanding of Verilog and some experience designing hardware (either ASIC or FPGA).
• A deep understanding of the issues involved in scheduling heavily pipelined architectures.
• A high-level understanding of the basic principles of compilers and compiler architecture.
• Some understanding of ML kernels and highly mathematical software.
• A strong understanding of the appropriate role of process in engineering development and the right time and ways in which to institute process during the ramp up of engineering development.

As Director of Hardware Abstraction, you will be chartered with developing the highest performance, easiest to use, simplest to learn, most effective, and fastest compile-time abstraction interface. For those who thrive on meeting challenges, Flex Logix provides an ideal supportive environment, intelligent personable colleagues, and plenty of interesting problems to challenge you.