Google settles with Singular Computing over claims of stolen AI chip tech

Search giant faced potential damages of over $5B but terms haven't been disclosed

Singular Computing's multi-billion dollar patent infringement lawsuit against Google was cut short on Wednesday after the search giant agreed to settle the case for an undisclosed sum.

A Massachusetts federal court filing [PDF] did not in fact reveal any terms of the settlement. However, pretrial documents [PDF] showed Singular was seeking between $1.6 billion and $5.2 billion in damages, over the alleged unlicensed use of its patents in the development of Google's tensor processing unit (TPU) v2 and v3 AI accelerators.

"We are pleased to have this matter resolved. We have always taken our disclosure obligations seriously and we will continue to do so," Google spokesperson José Castañeda told The Register.

Singular's attorney had not responded to a request for comment at the time of publication.

The settlement brings an end to the five-year legal spat between computer scientist Joseph Bates' company and Google. As we previously reported, between 2010 and 2014, Bates said he approached the Chocolate Factory and disclosed various patented technologies and prototypes he'd developed, under a non-disclosure agreement.

These patents (US 8,407,273, US 9,218,156, and 10,416,961) described a computer architecture designed to execute a large number of low-precision calculations each processor cycle. While impractical for conventional compute workloads Bates argued it was well suited to AI software, which could accommodate this low precision.

In its complaint, Singular accused Google of stealing Bates' designs and using them to develop into its second-and-third-gen TPU accelerators without permission or license.

Google asserted that no one who worked on its TPUs had any connection with Bates nor his blueprints. Internal emails, however, showed that Jeff Dean - now Google's chief scientist - had touted Bates' designs as being "really well suited" to the ad giant's workloads.

Throughout the suit Google repeatedly denied allegations of patent infringement, with the search biz previously telling The Register that Singular's patents "don't apply to our Tensor Processing Units, which we developed independently over many years."

Google’s chips, now in their fifth generation, are the backbone of its AI infrastructure and power a wide variety of the giant's internal machine learning workloads. Google makes the accelerators available to rent in its Cloud platform. ®

Send us news

Google thinks AI can Google better than you can

Multimodal models promise machines can do more – if you let them

Google I/O is Google A/I as search biz goes all-in on AI

Chrome is getting an onboard artificial intelligence model that devs can query

Google says its 'Trillium' TPUs are ready to power the next-generation of AI models

Promises 4.7x performance boost over its older silicon

Google Search results polluted by buggy AI-written code frustrate coders

Pulumi claims it has culled bad infrastructure-as-code samples

Big brains divided over training AI with more AI: Is model collapse inevitable?

Gosh, here's us thinking recursion was a solved problem

DeepMind spinoff Isomorphic claims AlphaFold 3 predicts bio-matter down to the DNA

AI may help drug discovery, but not US drug affordability

When AI helps you code, who owns the finished product?

It's not settled law. And it's going to mean trouble

Relax, Google's drop in search market share in April was just an illusion

Decline shown in data from StatCounter attributed to 'anomaly'

US senators' AI roadmap aims for $32B in R&D spending

'Reads like it was written by Sam Altman and Big Tech lobbyists,' say critics

Add AI servers to the list of iDevices Apple Silicon could soon power

Where have you been, Cupertino?

Investment analyst accuses Palantir of AI washing

Stick to data pipelines and ontology, says expert after share price dip

TikTok becomes first platform to require watermarking of AI content

The deepfake dystopia we’ve been waiting for has already arrived