It has been more or less confirmed that Intel will work with TSMC to realize its IDM 2.0 strategy. Ironic but true. The Santa Clara-based American chipmaker plans to achieve manufacturing leadership by the end of the decade. On this front, hundreds of billions of dollars have been allocated to expand wafer production capacity all over the world. Unfortunately for Intel, semiconductor foundries are a complex organism, requiring several years to reach full production capacity. This meant that an alternative source of cutting-edge chips needed to be secured to ensure leadership performance till the in-house foundries caught up. Thus began the romance between two of the largest foundry rivals, Intel and TSMC.
In the past, it has been reported multiple times that Intel would leverage TSMC’s 3nm process to manufacture its 3rd Gen Arc graphics cards, as well as some of its next-gen CPUs. The 6nm and 4nm nodes will be used to produce the 1st and 2nd Gen Arc GPUs, plus some chiplets for the Xe-HPC “Ponte Vecchio” lineup. Company CEO, Pat Gelsinger recently visited Taiwan to close this and some other future agreements.
A report from DigiTimes reveals the true nature of Intel and TSMC’s relationship. Although the latter demanded an advance payment to secure the latter’s 3nm wafer supply, it has invested a great deal to increase production capacity and ensure sufficient supply for the former’s leading chip families. In fact, TSMC has converted two factories (P8/P9) at Baoshan to fulfill Intel’s 3nm wafer needs. Each of these factories will produce 20K wafers per month in the first run of production with 40K slated for the second. These fabs will reportedly manufacture an important CPU family on TSMC’s 3nm node for Intel.
As reported earlier, Intel is likely to partner with TSMC to leverage its 2nm GAA process node as well. This technology will be used to manufacture its GPUs as well as certain CPUs in 2025. The 3nm Intel CPUs and GPUs are expected to hit retail towards the end of 2023, with 2nm coming sometime in 2025.
Intel coming in and suddenly splitting the most advanced wafers with Apple sounds rather far-fetched. But, this wouldn’t be the first time a major chipmaker got priority access to advanced technology using financial means. Gaining process leadership over AMD and NVIDIA may seem like a big win for Intel, but it’s worth noting that this will ensure sufficient supply for both without straining supply lines. AMD has been using TSMC’s 7nm node since 2020, and yet despite sticking to the same process technology has managed to bring significant improvements to IPC and boost clocks.
With TSMC’s N5P node slated to grant a 15-20% PPA boost to Zen 4, extracting another 10-15% from architectural advances won’t be too hard. Add advanced packaging technologies such as 3D V-Cache and hybrid core designs (read: Zen 4c), and this arrangement may just work in AMD’s favor as its greatest bottleneck through the pandemic has not been Intel, but limited supply.