January 20, 2022

We’re Utilizing an Optimized TSMC 5nm Course of


When AMD began utilizing TSMC’s 7nm course of for the Zen 2 processor household that launched in November 2019, one of many overriding messages of that launch was that it was essential to be on the vanguard of course of node know-how to be aggressive. That transfer to TSMC N7 was aided by the small chiplets used within the desktop processors on the time, making certain a better yield and higher binning curves for desktop and enterprise processors. Nevertheless, between at times, we’ve seen different firms make the most of TSMC’s 5nm, 4nm, and speak about TSMC’s 3nm course of coming to market over the subsequent 12-24 months. Throughout our roundtable dialogue with CEO Dr. Lisa Su, I requested if the necessity to keep on the vanguard nonetheless held true.

To place this into perspective, AMD introduced late in 2021 that it will be utilizing TSMC’s 5nm course of for its Zen 4 chiplets in enterprise CPUs within the second half of 2022. Then in early 2022, the corporate reiterated using Zen 4 chiplets, however this time in desktop processors once more by the tip of 2022. It is a important delay between the primary use of TSMC 5nm by the smartphone distributors, which reached mass manufacturing in Q3 2020, with Apple and Huawei being the primary to take benefit. Even right now, if we transcend 5nm, Mediatek has already introduced that its upcoming Dimensity 9000 smartphone chip is on TSMC 4nm and can come to market earlier this 12 months. TSMC’s 3nm course of is anticipated to ramp manufacturing on the finish of 2022, for a client launch in early 2023. By these metrics, AMD is behind a course of node or two by the point Zen 4 chiplets come to market later this 12 months.

I requested Dr. Su in our roundtable about whether or not the have to be on the vanguard course of is important to be aggressive for them. Having innovated round chiplets, I requested whether or not being the lead accomplice with foundry companions and packaging companions (often called OSATs) is of main significance, particularly when the lead competitors appear able to throw cash at TSMC to take that quantity. How would AMD have the ability to aggressively assert a market-leading place in gentle of the complexity of producing and the monetary energy of the competitors?

Dr. Su said that AMD is constant to innovate in all areas. For AMD it appears, main the chiplet know-how has helped to deliver the package deal collectively. She went on to say that AMD has had sturdy supply of 7nm, is introducing 6nm, adopted by Zen 4 and 5nm, speaking about 2D chiplets and 3D chiplets – AMD has all these items within the software chest and are utilizing the appropriate know-how for the appropriate utility. Dr Su strengthened that know-how roadmaps are all about making the appropriate selections and the appropriate junctures, and explicitly said that our 5nm know-how is extremely optimized for high-performance computing – it’s not essentially the identical as another 5nm applied sciences on the market.

Whereas not explicitly stating that the have to be vanguard is now not important, this messaging follows the improved narrative from AMD that within the period of chiplets, it’s how they’re mixed and packaged that’s turning into essential, arguably extra essential than precisely what course of node is getting used. We’ve seen this messaging earlier than from AMD’s principal competitor Intel, the place again in 2017 the corporate said that it’ll closely depend on optimized chiplets for every use case – this was crystallized additional in 2020 suggesting 24-36 chiplets on a single client desktop processor for purpose-built shopper designs. That being mentioned, it has been always rumored that Intel will probably be an enormous buyer of TSMC 3nm within the following years, so it will likely be fascinating to see the place AMD can make the most of a number of years of chiplet experience and packaging instruments by comparability.

Leave a Reply

Your email address will not be published. Required fields are marked *