TSMC won FinFET. All noteworthy leading edge logic designs, even Intel’s, are manufactured on TSMC’s N5 and N3 process in southern Taiwan. Competitors have been left in the dust. Samsung has had poor performance and poor yields since their 7nm, Intel is still early in its journey to recovery with Intel 4 and 3; neither have major external or internal customers ordering on these nodes in volume.
Future dominance is not guaranteed for TSMC. FinFETs, cannot scale further and SRAM shrink has been dead for a few nodes. The industry is at a crucial inflection point. Leading-edge logic must adopt two new paradigms in the next 2-3 years: gate all around (GAA) and backside power delivery (BSPDN or backside power delivery network).
Intel fell apart with their 10nm node and lost their 3 year lead for a number of reasons including not adopting EUV and transitioning to cobalt metallization with an immature tool supply chain despite warnings from Applied Material that their tools were not ready. The new paradigms of GAA and BSPDN present new opportunities in the foundry pecking order. They could even potentially open the door to a new entrant to the space – Rapidus, the Japanese government backed 2nm foundry startup.
As the capital expenditures needed to build leading edge fabs skyrockets, this means either Samsung or Intel could be forced to drop out of the race. Below we’ll discuss these topics in detail: a deep dive into BSPDN tech followed by leading edge logic roadmaps from all four foundries, competitiveness of their process technologies, SRAM scaling, and our model of 2nm capex and wafer fab equipment (WFE) spend.