The 7nm and 5nm races were almost single-handedly dominated by TSMC. Samsung’s 7nm node suffered from a lack of adopters, allegedly due to poor yields, while its 5nm node performs more or less the same as TSMC’s 7nm-class node. Furthermore, the Exynos 2100 which is fabbed on Samsung’s 5nm process is not only slower than the Qualcomm Snapdragon 888 (fabbed on TSMC’s 5nm EUV node) but also draws more power than it.
In the coming years, the competition will shift to 3nm-class process nodes, but before that, both foundries are expected to rely on their 4nm nodes (5nm/7nm derivative) to bridge the gap between contemporary and future technologies. Samsung’s 4nm LPP node is slated to be available to chipmakers sometime in 2022, the same as TSMC’s N4 process. The former will retain FinFET technology while the succeeding 3nm node will be based on GAA/GAE (Gate All Around/Game All Around Early) technology.
Rather surprising, Samsung is advertising its 4nm node as a major step on its process roadmap, bringing the same amount of power, performance, and density improvements as the 5nm node. Originally, its 4LPE node was marketed as an evolution of its 7LPP process. The reason being the improved use of EUV with 4nm as well as a notable improvement of around 50% in PPA.
Samsung is planning to start volume production of its 4LPE and 5LPP process nodes sometime in 2021, allowing consumers to opt for either depending on their PPA needs. While the 5nm LPE and LPP node mainly focus on transistor density and performance, the 4nm LPP node primarily improves power efficiency along with performance.
TSMC, on the other hand, is expected to start the risk production of its 4nm node in the second half of this year, with mass production slated for early 2022. Both Qualcomm and Apple are planning to take advantage of the foundry’s N4 node. The Snapdragon 895 and the Apple A16 Bionic will launch in mid and late 2022, respectively based on the same node.
The N4 node is going to feature a marginal increase in density compared to N5, with a density increase of around 6% thanks to optical shrink. The primary advantages, however, include lower process complexity, better performance, and easier porting. TSMC’s 4nm and 3nm nodes are both based on FinFET transistors, while Samsung is going to adopt GAA with its 3nm node.