Meta Seeks ASIC Designers For ML Accelerators And Datacenter SoCs

Meta wants to build accelerators and SoCs to run in its datacenters – for jobs including machine learning – but appears to be struggling to find folks to design them.

The Register has spotted numerous job ads for ASIC engineers with expertise in architecture, design, and testing in Bangalore, India, and Sunnyvale, California. Some of the jobs identify Meta as the employer, others indicate the work is for Facebook.

The architecture gigs detail the chance to "help architect state-of-the art machine learning accelerators."

Design gigs require candidates "with deep experience that spans one or more of the key areas required to build successful world-class complex SoC and IP for datacenter applications," and to do so for Facebook's Infrastructure organization in Bangalore, India.

Design verification gigs require "Experience with Design verification of Datacenter applications like Video, AI/ML and Networking designs."

Interestingly, some of the jobs were first posted to LinkedIn in late December 2023, but were re-posted two weeks ago. Some have just a handful of applicants, despite salaries approaching $200,000 for the jobs in Sunnyvale.

• Google releases Gemma – LLMs small enough to run on your computer
• In pursuit of artificial general intelligence, Meta adds Broadcom boss Hock Tan to its board
• Meta says risk of account theft after phone number recycling isn't its problem to solve
• OpenAI latest to add 'Made by AI' metadata to model work

Just what Meta wants to create isn't specified, but it's not hard to guess what The Social Network is up to – it's previously discusssed its homegrown "Meta Training Inference Accelerators" and its intention to bring them online in 2024.

With AI exploding and Nvidia admitting it will struggle to satisfy demand for its products, it would not surprise if Meta has decided to redouble its own efforts to ensure it doesn't have to bid for products in a hot, hot, market. Indeed, Nvidia CEO Jensen Huang on Wednesday claimed that his firm's hardware has helped to improve Meta's recommendation engines and by doing so improved its revenue. Meta therefore clearly has need for ML muscle, and may have decided to flex its cashflow to create it, rather than rely on a stressed supplier.

Meta is also pursuing artificial general intelligence – an effort that may require specialized silicon.

Whatever Meta wants to create, India's government will welcome Meta's decision to advertise in Bangalore. The nation sees its corps of semiconductor design talent as its bid to become a bigger player in the global semiconductor caper. New Delhi would be delighted by a designed-in-India ML accelerator.

We've asked Meta for info on what it plans to build but aren't holding our breath for a meaningful response. ®

PS: It was rumored this week – and we can well believe it – that Microsoft is developing a server networking card to improve the performance of machine-learning workloads and reduce its reliance on Nvidia and the network tech the GPU giant absorbed from its takeover of Mellanox.