IBM has sued GlobalFoundries for its failure to deliver a function 10nm node or the 7nm node that was supposed to replace it. The details in the case date back to 2014, when the two companies signed a deal for GlobalFoundries to build 10nm chips and POWER9 CPUs for IBM. IBM gave GlobalFoundries its Microelectronics division, and GlobalFoundries promised to support IBM for a decade. It was expected that GF would deliver 14nm on time, followed by a transition to 10nm.
Things didn’t go as planned. GlobalFoundries was late delivering 14nm and eventually canceled 10nm development altogether. IBM was apparently willing to work with GlobalFoundries on that issue, and would have accepted the company’s 7nm process as a substitute, had that node come to market. It didn’t. Instead, GlobalFoundries declared it would leave the leading edge and become a second-tier foundry, competing more with companies like UMC and SMIC as opposed to TSMC, Samsung, or Intel.
There’s an amusing quote in IBM’s court filing:
Those representations were an essential precondition to IBM entering into any relationship with GlobalFoundries, as GlobalFoundries had not been a significant player in the High Performance semiconductor business. Only by acquiring IBM’s technology, employees and intellectual property, would GlobalFoundries be able to expand into making High Performance semiconductor chips for IBM and for others.
That’s a bit much. The foundries GF bought — Fabs 9 and 10 — were outdated by the time it purchased them. Fab 9 is a 200mm facility that built 90nm and above chips. Fab 10 is newer, and currently builds 14nm chips with 300mm wafers, but it’s much smaller than the so-called “gigafabs” now deployed by Intel, TSMC, and Samsung. The idea that buying these two buildings was going to somehow rocket GlobalFoundries into a more competitive position doesn’t withstand the sniff test.
Aerial view of GlobalFoundries Malta campus, where its most advanced chips are built.11
What GlobalFoundries needed wasn’t IBM technology. It needed IBM’s business. One of the reasons given for why the company left the leading edge was its inability to land major customers. Ramping up a new node for 1-2 customers isn’t really feasible, especially if the chip volumes are relatively small. AMD had previously stated that it would no longer pay for custom, AMD-specific nodes. This left GlobalFoundries scrambling for customers. Had the company continued working on 7nm it would have benefited from the current semiconductor boom, but rumors at the time suggested GlobalFoundries’ owners were not willing to continue pouring money into a business that had never turned a profit.
Timothy Prickett Morgan at Next Platform has written an excellent story on the case, with a discussion of how the delay impacted the development of POWER9 and POWER10. He also provided a copy of IBM’s filing that wasn’t publicly available. GlobalFoundries’ response can be read here.
IBM’s complaint is straightforward: It signed a deal for 14nm and 10nm CPUs with GlobalFoundries back in 2015. Months later, GlobalFoundries declared it was canceling 10nm, but would offer IBM a 7nm process instead. It failed to do so. IBM declares:
Having taken the 1.5 billion and extracted the benefits of the best-in-class technology, engineers, and intellectual property from IBM, GlobalFoundries revealed that its representations about being committed to High Performance technology were false. GlobalFoundries misled IBM in order to acTuire IBM’s semiconductor business and form an ostensibly mutually beneficial alliance, which GlobalFoundries then deliberately and dishonorably discarded when it suited GlobalFoundries’ purposes.
GlobalFoundries Response
GlobalFoundries’ response to IBM’s accusations of greed boil down to “no u” with a side order of suspicious timing. GF argues that suit has been brought at a time when it is preparing for an IPO. A company with a large lawsuit hanging over its head from IBM might not be valued as highly as it would be otherwise. GF notes that it stopped developing 10nm roughly 6 years ago and stopped developing 7nm approximately three years ago. It writes:
IBM watched GF’s success and the news of the IPO, then seeking a quick payday in the hopes that GF would not defend itself, IBM sent a legal dispute letter from a top tier, expensive white-shoe New York law firm to GF, demanding that it pay an outlandish $2.5 billion in unspecified damages for alleged breaches that, until April 2021, had never been asserted, and without any real explanation as to the basis of its claims… IBM also has a reputation of making licensing requests, at or near, quarter end to meet Wall Street analyst earning expectations and this aggressive move is consistent with that playbook.
GF also claims that it sought to remedy this problem through an agreed-upon, 60-day dispute process, but received no response from IBM. The company seeks a declaration that it did not violate the letter of its agreement with IBM and that the company’s lawsuits are not based on any broken contract. It also argues that IBM is upset that semiconductor demand has spiked some years after it sold its fabs.
There is one seeming point of agreement between the two companies. IBM notes in its suit that GF never intended to become a leading producer of 7nm chips in terms of total volume. GlobalFoundries acknowledges this tacitly when it writes:
IBM quickly found a new, and less expensive, 7nm technology supplier in Samsung. Indeed, the arrangement worked out better for IBM as Samsung, upon information and belief, was able to supply the technology more quickly than GF would have been able to and at significantly less cost to IBM. IBM actually benefitted from GF’s decision to cease working on 7nm technology.
Arguing that your partner benefited from your own inability to build microprocessors is a bold move. We’ll see if it pays off for them.
Feature image shows GlobalFoundries’ Dresden facility from the air. Image by Fensterblick, CC BY-SA 3.0.
Now Read:
0 Commentaires