Soheil Hashemi

Hardware Design

- 10+ years in ASIC design, approximate computing, energy-efficient deep learning hardware
- Ph.D. ECE Brown University

Why Fathom

The future will be the era of artificial intelligence, and AI relies on computing models with millions of parameters and connections. Still, these models are constantly increasing in size with more and more connections, demanding higher compute power. Yet, more compute FLOPS does not translate to performance unless there is an interconnect feeding the compute logic just as fast. While historically the interconnect has not been scaling nearly at the same rate as the compute, few companies, let alone startups, have risen up to the challenge of interconnect. Simply put, conventional approaches to the interconnect are still limited by the 2D limitations of area and distance.

Enter Fathom Computing. Fathom’s innovative free space optics approach, aims at solving this bottleneck by adding a third dimension to the equation, enabling millions of parallel low-power channels. The non-blocking nature of such an all-to-all interconnect results in groundbreaking bandwidths delivered to the compute.

Fathom has assembled a dynamic team of interdisciplinary scientists and engineers, from optics to hardware design and ML, working on making Fathom’s vision a reality. As someone who enjoys building computing system hardware, I want to contribute to this vision, bringing on the next revolution in computing.