Microsoft offers electrical engineers a lifeline as it pursues custom cloud silicon
Redmond see, Redmond do... what AWS and Google are also doing
As custom silicon plays an ever-larger role in the cloud, Microsoft is more and more openly gearing up to design its own homegrown datacenter-grade chips.
It's long been rumored and whispered that the Windows giant is working on its own processors for servers as well as personal devices. Google and Amazon are doing the same for their clouds, and even Meta has impressive chips of its own.
What we have here is more evidence of Microsoft's drive to design its own silicon the way it wants for its own specific purposes, creating custom components to run and accelerate workloads rather than use processors from someone else. And that evidence is in the form of a slew of fresh Microsoft Silicon job ads seeking engineers for various projects, including at least one to build a custom AI accelerator.
This follows on from previous ads from Redmond for chip designers. It's understood these latest job postings relate to the corporation's Azure cloud.
"We are looking for a principal design engineer to work in the dynamic Microsoft Artificial Intelligence System on Chip Silicon team," a listing posted this very week reads.
That team is said to be working on "cutting-edge AI designs that can perform complex and high-performance functions in an extremely efficient manner." In other words, having hitched its future on OpenAI's family of technologies, Microsoft wants to craft chips to run those models that are more efficient than off-the-shelf GPUs and related accelerators.
From the listings, it's clear Microsoft is looking for experienced engineers. The aforementioned position calls for an electrical engineer with at least nine years of time in the industry who will be responsible for logic design and micro-architectural work for high-performance components. The base pay tops out at $280,000 a year.
Microsoft's interest in custom silicon extends beyond AI accelerators. A separate posting from Redmond's Silicon Computing Development Organization is seeking a design verification engineer to serve on its pre-silicon hardware verification team. According to the posting, that team will be looking at SoC designs for cloud workloads, suggesting Microsoft is considering a custom processor along the lines of Amazon's Graviton family.
Microsoft has also posted openings for a silicon engineer to work on data processing units (DPU) and a packaging design engineer within Azure's Hardware Systems and Infrastructure Group.
DPUs — sometimes called smartNICs or Infrastructure Processing Units — offload various functions, such as security, networking, or storage, from the host CPU cores. This is perhaps the least surprising listing as Microsoft acquired DPU vendor Fungible in January.
The latter position, meanwhile, would be responsible for "delivering advanced packaging solutions for HPC silicon designs" for "various datacenter product segments." This suggests that Microsoft aims to follow in Amazon's footsteps and build custom silicon for all manner of compute applications.
- Ampere heads off Intel, AMD's cloud-optimized CPUs with a 192-core Arm chip
- AMD reveals Azure is offering its SmartNICs as-a-service
- Google IO: A deeper dive into the developer day's details
- Broadcom chases AI craze with ML-tuned switch ASICs
The full scope of Microsoft's silicon ambitions aren't clear, and just because the company is hiring chipset engineers doesn't mean we'll necessarily see custom parts anytime soon. All Microsoft's representatives could tell us is this:
We’re continuing to invest in our own capabilities and fostering and strengthening partnerships with a wide range of chip/ecosystem providers. Our goal is to provide rich solutions for our customers with a systems-wide approach.
Pursuing custom silicon certainly has its advantages. In addition to cutting out the middleman, custom chips present an opportunity to build processors for domain-specific applications that are more efficient than off the shelf parts.
Both Amazon Web Services and Google Cloud have developed custom AI accelerators. Amazon has Trainium and Inferentia, and Google has their Tensor Processing Units (TPUs) now in their fourth generation. By comparison, Microsoft has largely relied on off the shelf or customized hardware from chipmakers like Nvidia, AMD, and Intel, which means it has a lot of catching up to do.
However, the past few quarters have seen mass layoffs across much of the semiconductor ecosystem, which could present an opportunity for Microsoft to fill the gaps in its talent pool. ®