The semiconductor shortages have made chip manufacturing facilities more important than ever. In this regard, TSMC is easily the most respected and sought after. From Apple to Intel, AMD, Qualcomm, MediaTek, etc, more than half of all processors are manufactured at one of TSMC’s fabs. To nullify this advantage (to an extent), Intel will be playing both sides in the next round of foundry wars: Using a consistent supply of TSMC’s latest and greatest wafers, while also working to fix its internal fabs.
The Taiwanese foundry’s next-gen node 3nm (N3) is slated to begin volume production in the last months of 2022. With a performance uplift of 15%, a power efficiency gain of 30%, and density improvement of 70% over the existing 5nm (N5) node, it is expected to gain widespread adoption by 2024.
Some clients, however, will get priority access to the advanced node. Till now, this privilege has been primarily enjoyed by Apple. With the 3nm node, Intel will be joining that club, splitting the first wave of 3nm shipments evenly with the Cupertino giant. These wafers will be used to manufacture the 3rd/4th Gen of Arc GPUs, codenamed Celestial and Druid. Although there’s limited evidence, some rumors have indicated that the 14th and 15th Gen Core processors will also leverage iGPU chiplets fabbed on a TSMC process.
This will give Intel a full cycle node advantage over AMD and NVIDIA, at least in the GPU market. Both AMD and NVIDIA will transition to the 3nm node no sooner than early 2024. This is when AMD will launch its Zen 5-based processors which will reportedly mix and match cores and process nodes in one design. NVIDIA will also launch the successor to Ada Lovelace (RTX 40 series) later that year in the form of the RTX 50 series (possibly Blackwell) on the same node.
On the bright side, this means that AMD won’t be starved of its supply of 5nm/6nm wafers as its larger rivals race to the 3nm node. The chipmaker has reportedly procured more than enough capacity to ensure a consistent supply of its next-gen Zen 4 and RDNA 3 processors. Both these families will be fabbed on the 5nm process node, with a launch targeted for late 2022.
Using a mix of advanced packaging technologies and architectural prowess, AMD might as well come out on top despite lagging behind on the node front. The 3D V-Cache stacking on the Zen 3D/Zen 4 CPU side, and a chiplet based GPU approach are bound to make waves in their respective markets.