AMD has tapped into Samsung’s 4nm node capacity to mass produce its Zen 5c dies, marking a shift in its single-foundry strategy. Team Red will split its Zen 5c production between TSMC and Samsung, using the latter for the “basic” variants of the Prometheus (codename for Zen 5c) die. According to the report from DigiTimes, TSMC will continue to manufacture the advanced variant of the Zen 5c die on its 3nm-class process node.
TSMC has been AMD’s primary foundry partner since the introduction of the Zen architecture in 2019, leaving Global Foundries for the less-important I/O and SoC nodes. If this report holds, it will be the chipmaker’s first attempt at diversifying its supply chain. Rival NVIDIA has followed a dual foundry approach for quite some time, though its most recent and upcoming nodes are expected to leverage a cutting-edge TSMC node.
Generally, chipmakers tend to utilize the same process technology for a given IP. Then again, AMD did manufacture RDNA 3 GPUs across different process nodes (5nm for Navi 31/32 and 6nm for Navi 33).
Apple is the only client to utilize TSMC’s 3nm capacity. As per rumors, the client-class Zen 5 chips will be fabbed on the older 5nm-class node, leaving the lucrative server segment to the 3nm capacity. It is possible that AMD decided to tap into Samsung’s 4nm node to diversify its supply while retaining the performance/efficiency advantages.
Samsung’s 3nm node leverages the GAA (Gate All Around) transistors, an evolution of the FinFET technology used by TSMC’s 3nm nodes. Unfortunately, the yields of advanced nodes have been on the lower end, pushing most fabless chipmakers toward TSMC.