When asked about Chatterjee, Google spokesman Jason Freidenfelds provided a statement from the company confirming that he was “terminated for cause”. Freidenfelds also provided a statement from Zoubin Ghahramani, vice president of Google Research, saying that “we are determined to uphold our standard of respectful discourse among our researchers.” Gharamani’s statement did not mention Chatterjee by name.
The episode adds to a series of recent internal conflicts at Google that showcase the liberal, engineer-centric culture it celebrates as a startup. leaving the company unprepared for some of the challenges of being a multinational company with over 100,000 employees.
Google hired Satrajit Chatterjee in 2018 as a senior machine learning researcher. Previously, he was senior vice president of hedge fund Two Sigma and also worked at Intel. When Chatterjee joined, Mirhoseini and Goldie were already working in the company’s most prominent machine learning lab, Google Brain. Chatterjee joins a separate, smaller research group within Google’s research division.
The two women did not work directly with Chatterjee, but in 2019, Goldie’s internal documents claimed, he asked for the Morpheus project manager. After being politely turned down, staff say, Chatterjee began to doubt the pair’s work with senior researchers they needed to collaborate with or win support with, showing the Their work is false or even fabricated.
As a senior employee, Chatterjee’s questions can be influential. As a result, employees said, other senior employees sometimes cast doubt on Goldie and Miroseini’s work, questioning their results.
The effect was to turn Miroseini and Goldie’s work at Google into a tense, divisive reality, insiders assert. At the same time as running a successful project with support from Google’s chip designers, they say the pair had to do extra work to respond to accusations that their results were false or even wrong.
Chip design teams at Google and elsewhere are generally very cautious, because making nanochips is expensive and any defect in the chip cannot be corrected once it has been etched into silicon. Google has said that TPU has made breakthroughs in its AI research and serviceand lease chips through its cloud unit. However, Chatterjee’s criticisms of Morpheus continued even after Google’s hardware leaders decided they trusted it enough to help it design the company’s next generation of TPUs.
In May 2021, a Google employee posted to an internal email list asking if anyone had applied machine learning technology to circuit board design. Mirhoseini replies that Morpheus can help. But Chatterjee concurs in claiming that older techniques better than commercially available machine learning and chip design tools have yielded the best results.
Jeff Dean, Google’s head of AI, joined the discussion to say that Morpheus was used to design the next generation of TPU chips. The technology has won large-scale tests against human chip experts and commercial chip design tools, Dean said, and attached a slide show of the results.
Dean also linked to the team’s recently published paper, peer-reviewed Nature research. It reports that the Morpheus team’s code produces better TPU circuit blocks than Google engineers using commercial chip design tools. The authors did not disclose details about those chip segments, saying they were confidential to Google, but also included the results for an open-source processor design that is freely available to anyone. The results of the paper were then copied by another research team inside Google, and the code for the tests was open source.
