Samsung to Unveil Refined 3nm and Efficiency-Enhanced 4nm Nodes at VLSI Symposium
Samsung Foundry is ready to element its second technology 3 nm-class fabrication expertise in addition to its performance-enhanced 4 nm-class manufacturing course of on the upcoming upcoming 2023 Symposium on VLSI Technology and Circuits in Kyoto, Japan. Each applied sciences are necessary for the contract maker of chips as SF3 (3GAP) guarantees to supply tangible enhancements for cell and SoCs, whereas SF4X (N4HPC) is designed particularly for probably the most demanding high-performance computing (HPC) purposes.
2nd Era 3 nm Node with GAA Transistors
Samsung’s upcoming SF3 (3GAP) course of expertise is an enhanced model of the corporate’s SF3E (3GAE) fabrication course of, and depends on its second-generation gate-all-around transistors – which the corporate calls Multi-Bridge-Channel field-effect transistors (MBCFETs). The node guarantees extra course of optimizations, although the foundry prefers to not examine SF3 with SF3E. In comparison with its direct predecessor, SF4 (4LPP, 4nm-class, low energy plus), SF3 claims a 22% efficiency enhance on the identical energy and complexity or a 34% energy discount on the identical clocks and transistor rely, in addition to a 21% logic space discount. Although it’s unclear whether or not the corporate has achieved any scaling for SRAM and analogue circuits.
As well as, Samsung claims that SF3 will present extra design flexibility facilitated by various nanosheet (NS) channel widths of the MBCFET gadget throughout the identical cell kind. Curiously, variable channel width is a characteristic of GAA transistors that has been mentioned for years, so the way in which Samsung is phrasing it in context of SF3 may imply that SF3E doesn’t help it.
To this point neither Samsung LSI, the conglomerate’s chip growth arm, nor different clients of Samsung Foundry have formally launched a single highly-complex processor mass produced on SF3E/3GAE course of expertise. In actual fact, it seems like the one publicly-acknowledged utility that makes use of the business’s first 3 nm-class fabrication course of is a cryptocurrency mining chip, in line with TrendForce. This isn’t notably stunning as utilization of Samsung’s ‘early’ nodes is usually fairly restricted.
Against this, Samsung’s ‘plus’ applied sciences are sometimes utilized by a variety of shoppers, so the corporate’s SF3 (3GAP) course of is prone to see a lot increased volumes when it turns into obtainable someday in 2024.
SF4X for Extremely-Excessive-Efficiency Purposes
Along with SF3, which is designed for a wide range of attainable use circumstances, Samsung Foundry is prepping its SF4X (4HPC, 4 nm-class high-performance computing) designed for performance-demanding purposes like datacenter-oriented CPUs and GPUs.
To handle such chips, Samsung’s SF4X affords a efficiency enhance of 10% coupled with a 23% energy discount. Samsung would not explicitly specify what course of node that comparability is being made in opposition to, however presumably, that is in opposition to their default SF4 (4LPP) fabrication expertise. To attain this, Samsung redesigned transistors’ supply and drain after reassessing their stresses (presumably beneath excessive masses), carried out additional transistor-level design-technology co-optimization (T-DTCO), and launched a brand new middle-of-line (MOL) scheme.
The brand new MOL enabled SF4X to supply a silicon-proven CPU minimal voltage (Vmin) of 60mV, a ten% lower within the variation of off-state present (IDDQ), assured excessive voltage (Vdd) operation at over 1V with out efficiency degradation, and an improved SRAM course of margin.
Samsung’s SF4X might be a rival for TSMC’s N4P and N4X nodes, that are due in 2024 and 2025 respectively. Primarily based on declare specificaitons alone, it’s onerous to inform which expertise will provide the most effective mixture of efficiency, energy, transistor density, effectivity, and price. That mentioned, SF4X might be Samsung’s first node within the current years that was particularly architected with HPC in thoughts, which suggests that Samsung has (or is anticipating) sufficient buyer demand to make it value their time.
