The large image: Samsung’s upcoming 3nm-class lithography course of might be their first to make use of gate-all-around (GAA) transistor know-how. Within the face of considerations raised about if the corporate can have GAA designs out in time to maintain tempo with competitor TSMC, it insists that its first iteration of the know-how will start mass manufacturing in 2022.
Samsung’s public roadmap was offered just lately at its Foundry Discussion board 2021 in China. Significantly conspicuous in its absence from the slide proven, nonetheless, was Samsung’s was its 3GAE (3nm, GAA-Early), the primary iteration of 3nm know-how. 3GAE was initially revealed in 2019 alongside its follow-up, 3GAP (3nm, GAA-Plus), however solely the latter was proven off within the presentation.
Previous to this roadmap being revealed, there had already been concern brewing over the well being of the node; 3GAE had initially been deliberate for threat manufacturing in late 2020 and quantity manufacturing in 2021, however the firm solely taped out its first 3nm take a look at chips final month.
Moreover, Dr. Chidi Chidambaram, VP of Engineering at Qualcomm, Samsung Foundry’s largest third-party buyer was quoted by SemiAnalysis as estimating GAA know-how as solely reaching manufacturing in 2023-24, at a latest occasion hosted by Utilized Supplies.
Whereas the estimate is probably going deliberately imprecise for NDA causes, it nonetheless places Samsung a yr behind TSMC’s plans for quantity manufacturing of 3nm-class silicon subsequent yr. This, mixed with the absence of 3GAE on the printed roadmap, led to hypothesis that the node had been utterly left out in favor of 3GAP.
Simply to elucidate the naming by Samsung: 3GAE (Gate-All-Round Early) is lacking right here. This means that the early course of is just not viable for mass manufacturing and bought killed utterly. 3GAP (Gate-All-Round Plus) now planed for HVM in 2023, which is confirming Qualcomm’s assertion https://t.co/9GeMcLq3jt
— Andreas Schilling (@aschilling) July 7, 2021
When contacted for remark by Anandtech, the corporate mentioned that 3GAE was nonetheless due for mass manufacturing in 2022, and it is attainable that the node is not being publicized just because it is reserved for inside utilization at Samsung LSI.
Nonetheless, whereas this was achieved with different “-Early” course of variants from the corporate previously, these had been nonetheless marked on the map; 5LPE noticed third-party utilization in Qualcomm’s Snapdragon 888, however that node was itself derived from 7LPP.
The corporate can be placing out a 4LPP, primarily based on extra conventional FinFET know-how. This (and its predecessor 4LPE) at the moment are listed as their very own household of nodes, as a substitute of being an evolution from 5nm/7nm-class applied sciences.
It might be that the method has a considerable sufficient enchancment to warrant advertising and marketing as a brand new “headline” node, or just due to extra vital variations in design and manufacturing.
Nonetheless, the absence of 3GAE on the general public roadmap and the FinFET-based 4LPP node overlapping with Samsung’s said 2022 goal for 3GAE attracts uncomfortable comparisons with Cannon Lake, Intel’s earliest 10nm processors.
These technically shipped in 2018 to satisfy investor commitments, however had been so unhealthy that they had been handed over in favor of additional evolutions of the older 14nm node; the corporate now ignores that technology utterly, insisting that Ice Lake was its first actual 10nm line-up.
However, maybe, no less than the oft-mocked 14+++ was extra simply comprehensible than a branching tree of processes and their sub-evolutions.