Digitimes: TSMC 3nm reportedly suffering from bad yields even as the company defiantly claims mass production will begin in Q2 2022
AMD is already one of the biggest customers when it comes to TSMC and arguably the most loyal (in the PC hardware space). Intel has its own foundry and is trying to actively compete with TSMC. NVIDIA has dilly-dallied with Samsung in the past but will return to TSMC for its 5nm GPUs after having paid billions to secure allocation in a foundry ecosystem that is basically running at a 100%. With that context, TSMC remains the most important pure-play foundry in the world and it is understandable that they will say absolutely nothing about any delays unless it is unavoidable. The company has already rolled out a version of the 3nm process called the N3E, which will be a lighter, lower-power version of the full-fledged N3 process. Considering Intel is set to introduce its Intel 4 process in Q2 0222, such a delay could not only put it back on the map as a leading-edge foundry but also allow it to offer its foundry services to other customers. Considering the deteriorating geopolitical environment with regards to Taiwan and Intel being the likely recipient of billions of dollars of subsidies under the CHIPS Act, it could prove to be an opportunity for the blue-chip maker, served on a silver platter. It is worth noting that while the nomenclature for Intel’s process is simply Intel 7 to Intel 4 in Q2 2022, this constitutes a major node jump for the company. Intel 7 is the former Intel 10nm process (which was comparable in fin and pitch size to TSMC 7nm so renamed) while Intel 4 represents the original 7nm process. The company is also investing heavily in high-NA lithography from ASML and remains in an excellent position for a comeback in a year or two. If TSMC does indeed stumble, then, AMD and NVIDIA will be forced to switch to N4 (4nm process), which would allow Intel to start clawing back market share from AMD and could even dent NVIDIA’s GPU market (assuming it starts to fabricate ARC GPUs in-house (they are currently being made on TSMC’s 6nm process). All of this is - assuming - Intel does not stumble again as it did with the original 10nm process. Something that caused it to go from a foundry comfortably a few years ahead of TSMC to lagging behind.
