https://fuse.wikichip.org/news/7343/iedm-2022-did-we-just-witness-the-death-of-sram/ Skip to content Wednesday, December 14, 2022 Latest: * IEDM 2022: Did We Just Witness The Death Of SRAM? * SiFive Announces New Cores, BiFurcates Line Into Performance And Efficiency Cores * Intel, SiFive Demo High-Performance RISC-V Horse Creek Dev Platform On Intel 4 Process * TSMC Demos SoIC_H for High-Bandwidth HPC Applications * Intel Rolls Out 13th Gen Core, Raptor Lake Processors; Cranks Up The Frequency * WikiChip Fuse WikiChip Fuse Your Chips and Semi News [ ] * Home * Account * Main Site * Architectures + x86 + ARM + RISC-V + Power ISA + MIPS * Supercomputers * 14 nm * 12nm * 10nm * 7nm * 5nm Foundries IEDM 2022: Did We Just Witness The Death Of SRAM? December 14, 2022December 14, 2022 David Schor 3 nm, cache, N3, N3B, N3E, SRAM, TSMC [iedm-n3-present-header] [iedm-logo-68-300x146] This year, the 68th Annual IEEE International Electron Devices Meeting (IEDM) was back in full force with nearly 1,500 engineerings (in-person) from around the world making their yearly voyage back to downtown San Francisco to discuss the latest development in the semiconductor industry. While there were a great number of interesting papers from both academia and industry, it was the one by TSMC that brought frighteningly bad news - whereas logic is still scaling more-or-less along the historical trendline, SRAM scaling appears to have completely collapsed. No Scaling! At the conference, TSMC talked about both the original base N3 (N3B) node as well as the enhanced variant (N3E) which is a slightly more relaxed variant of the N3B. What's interesting is, for the new N3E node, the high-density SRAM bitcell size as not shrunk at all. Coming in at 0.021 um2, this is exactly the same bitcell size as their N5 node. The N3B variant which isn't expected to make it into too many products did have a scaled SRAM bitcell; however, at 0.0199um2, it's a mere 5% scaling (or 0.95x shrink). [sram-density-tsmc-n3b-n3e] In terms of rough memory density (assuming ISO-assist circuit overhead), N3E is roughly at 31.8 Mib/mm2 and would increase to 33.55 Mib/mm2 or 1.75 Mib/mm2 (230 kilobytes) improvement. Impact That's some seriously bad news! To put that into perspective, while N3B and N3E are both said to deliver a 1.6x and 1.7x chip-level transistor scaling, a 1.0x and 1.05x scaling of SRAM is catastrophic. Now, we still expect TSMC to roll out a denser SRAM bitcell variant for N3 at some point and we do expect to see some level of shrinkage for SRAM in the future, but the good old shrink SRAM shrinkage appears to be dead. Consider a hypothetical 10-billion transistor chip comprising 40% SRAM and 60% logic on TSMC N16. Ignoring practical limitations and analog/phys/etc, such a hypothetical chip would be around 255 mm2 with 45 mm2 or 17.6% accounted for SRAM. Shrinking the exact same chip down to N5 would produce a 56 mm2 chip with 12.58 mm2 or 22.5% of the die accounted for SRAM. Shrinking the chip further to N3 (based on our initial - but not fully confirmed - values) would produce a 44 mm2 chip with the same 12.58 mm2 SRAM density which now accounts for nearly 30% of the area. [n16-n5-n3-scaling-chip] Of course, the impact won't be felt the same across the board. The percentage of SRAM and caches on chips varies greatly by your target market and overall capabilities. However, for some AI hardware startups where the architecture dictates that a large portion of the chip be covered by SRAM, those engineers will encounter more challenges quicker than others. Industry-Wide The collapse of SRAM scaling is not restricted to just TSMC. We have pointed out the slowing SRAM scaling issue for quite some time now. For example, while Intel is still shrinking its SRAM bitcells, the company's most recently-announced Intel 4 process SRAM scaling slowed down to 0.7-0.8x scaling from their historical 0.5-0.6x scaling. For Intel 4, our estimated density (at ISO-assist circuit overhead vs TSMC) is at 27.8 Mib/mm2 or 4 Mib/mm2 or 13% behind. It would not be unrealistic to expect Intel's Intel 3 process to match or beat them. [sram-density-tsmc-n3be] So where do we go from here? The reality is that currently, the only viable alternative to SRAM is simply more SRAM, and therefore we expect to see SRAM directly consuming more area. That's not to say we don't expect more SRAM scaling. While we do expect TSMC and other foundries to roll denser SRAM, it would appear that the historical scaling is officially dead. Some research institutions such as Imec presented higher-density SRAM bitcells. For example last year, at IEDM 2021, Imec presented SRAM densities of around 60 Mib/mm2 or roughly twice today's densities on a hypothetical "beyond 2 nm node" utilizing forksheet transistors and an advanced dual-size interconnect scheme. Beyond SRAM, the industry has been looking into many other alternative memory architectures. Those emerging memory bitcells offer unique tradeoffs when compared to SRAM such as higher density at lower read/write specifications, non-volatility capabilities, lower read-write cycle capabilities, or lower power at potentially lower density or speeds. While they are not direct replacements for SRAM, moving forward they might play a role as level 4 or level 5 caches where the lower performance tradeoffs can be offset by higher density. For now, it seems that the industry has reached an interesting inflection point. --------------------------------------------------------------------- Further Reading: * N3E Replaces N3; Comes In Many Flavors * A Look At Intel 4 Process Technology * TSMC 2021 Foundry Update: Foundry Roadmap * TSMC 2021 Foundry Update: Automotive, Networking, and HPC Roadmap --------------------------------------------------------------------- - Spotted an error? Help us fix it! Simply select the problematic text and press Ctrl+Enter to notify us. - * - SiFive Announces New Cores, BiFurcates Line Into Performance And Efficiency Cores * Share This Post: Related Articles TSMC 2021 Foundry Update: Foundry Roadmap TSMC Extends Its 5nm Family With A New Enhanced-Performance N4P Node Q1 2021 Foundry Update: Spending Bonanza Intel 2020s Process Technology Roadmap: 10nm+++, 3nm, 2nm, and 1.4nm for 2029 AMD 3D Stacks SRAM Bumplessly QUEST, A TCI-Based 3D-Stacked SRAM Neural Processor Top Six Articles * IEDM 2022: Did We Just Witness The Death Of SRAM? * TSMC Details 5 nm * N3E Replaces N3; Comes In Many Flavors * TSMC Talks 7nm, 5nm, Yield, And Next-Gen 5G And HPC Packaging * Intel 2021 Process Technology Update: Intel 7, Intel 4, Intel 3, and Intel 20A * TSMC 7nm HD and HP Cells, 2nd Gen 7nm, And The Snapdragon 855 DTCO Ezoicreport this ad Recent * IEDM 2022: Did We Just Witness The Death Of SRAM? IEDM 2022: Did We Just Witness The Death Of SRAM? December 14, 2022December 14, 2022 David Schor * SiFive Announces New Cores, BiFurcates Line Into Performance And Efficiency Cores SiFive Announces New Cores, BiFurcates Line Into Performance And Efficiency Cores November 1, 2022November 2, 2022 David Schor * Intel, SiFive Demo High-Performance RISC-V Horse Creek Dev Platform On Intel 4 Process Intel, SiFive Demo High-Performance RISC-V Horse Creek Dev Platform On Intel 4 Process October 7, 2022October 7, 2022 David Schor * TSMC Demos SoIC_H for High-Bandwidth HPC Applications TSMC Demos SoIC_H for High-Bandwidth HPC Applications October 4, 2022October 4, 2022 David Schor * Intel Rolls Out 13th Gen Core, Raptor Lake Processors; Cranks Up The Frequency Intel Rolls Out 13th Gen Core, Raptor Lake Processors; Cranks Up The Frequency October 1, 2022October 3, 2022 David Schor * The NPU Inside Every Intel PC The NPU Inside Every Intel PC September 30, 2022September 30, 2022 David Schor Ezoicreport this ad Random Picks Samsung Foundry On EUV, Pellicles, Capacity, and Yield Samsung Foundry On EUV, Pellicles, Capacity, and Yield July 25, 2022July 25, 2022 David Schor Eni fires up its supercomputer, breaks into the TOP500's top ten Eni fires up its supercomputer, breaks into the TOP500's top ten January 19, 2018May 25, 2021 David Schor Cavium Takes ARM to Petascale with Astra Cavium Takes ARM to Petascale with Astra August 25, 2018May 25, 2021 David Schor Cambricon Reaches for the Cloud With a Custom AI Accelerator, Talks 7nm IPs Cambricon Reaches for the Cloud With a Custom AI Accelerator, Talks 7nm IPs May 26, 2018May 25, 2021 David Schor ARM Announces the Mali-C52 and Mali-C32 ISPs for Intelligent Vision Devices ARM Announces the Mali-C52 and Mali-C32 ISPs for Intelligent Vision Devices January 3, 2019May 25, 2021 David Schor Random Tags 2.5D packaging 3 nm 3D packaging 5 nm 5nm 7 nm 7nm 10 nm 10nm 14 nm 16nm AI AMD ARM ARMv8 ARMv9 chiplet Coffee Lake Core i5 Core i7 Cortex edge computing EMIB EUV FinFET GlobalFoundries Hot Chips IBM Ice Lake IEDM inference Intel Intel 7 ISSCC multi-chip package neural processors RISC-V Samsung subscriber only (general) Sunny Cove Supercomputers TSMC VLSI Symposium x86 Zen x86 WorldView All Intel Introduces Thread Director For Heterogeneous Multi-Core Workload Scheduling Desktop Processors Mobile Processors Intel Introduces Thread Director For Heterogeneous Multi-Core Workload Scheduling August 19, 2021August 19, 2021 David Schor Intel introduces the Intel Thread Director for heterogeneous multi-core workload scheduling Intel Unveils Sapphire Rapids: Next-Generation Server CPUs Architectures Server Processors Intel Unveils Sapphire Rapids: Next-Generation Server CPUs August 19, 2021August 19, 2021 David Schor Intel's Gracemont Small Core Eclipses Last-Gen Big Core Performance Architectures Data Processing Unit Desktop Processors Mobile Processors Intel's Gracemont Small Core Eclipses Last-Gen Big Core Performance August 19, 2021August 21, 2021 David Schor Intel Unveils Alder Lake: Next-Generation Mainstream Heterogeneous Multi-Core SoC Architectures Desktop Processors Mobile Processors Intel Unveils Alder Lake: Next-Generation Mainstream Heterogeneous Multi-Core SoC August 19, 2021August 19, 2021 David Schor Intel Details Golden Cove: Next-Generation Big Core For Client and Server SoCs Architectures Desktop Processors Mobile Processors Server Processors Intel Details Golden Cove: Next-Generation Big Core For Client and Server SoCs August 19, 2021August 19, 2021 David Schor Intel Launches 3rd Gen Ice Lake Xeon Scalable Architectures Server Processors Intel Launches 3rd Gen Ice Lake Xeon Scalable April 6, 2021May 23, 2021 David Schor Random MediaTek Announces The Helio P22, A New Premium Mid-Range SoC MediaTek Announces The Helio P22, A New Premium Mid-Range SoC May 25, 2018May 25, 2021 Matt Larson AMD Announces Threadripper 2, Chiplets Aid Core Scaling AMD Announces Threadripper 2, Chiplets Aid Core Scaling August 7, 2018May 25, 2021 David Schor Samsung 2nd gen 10nm enters HVM, S3 fab ready for primetime Samsung 2nd gen 10nm enters HVM, S3 fab ready for primetime November 29, 2017May 25, 2021 David Schor QUEST, A TCI-Based 3D-Stacked SRAM Neural Processor QUEST, A TCI-Based 3D-Stacked SRAM Neural Processor April 25, 2018May 25, 2021 David Schor IEDM 2017 + ISSCC 2018: Intel's 10nm, switching to cobalt interconnects IEDM 2017 + ISSCC 2018: Intel's 10nm, switching to cobalt interconnects February 17, 2018May 25, 2021 David Schor IBM Releases Power ISA v3.1; To Present POWER10 At Hot Chips 32 IBM Releases Power ISA v3.1; To Present POWER10 At Hot Chips 32 May 23, 2020May 23, 2021 David Schor Intel Updates Apollo Lake: More LPC Reliability Issues Intel Updates Apollo Lake: More LPC Reliability Issues September 9, 2019May 25, 2021 David Schor ARM WorldView All Arm Refreshes The Cortex-A510, Squeezes Higher Efficiency Architectures Mobile Processors Arm Refreshes The Cortex-A510, Squeezes Higher Efficiency June 28, 2022June 28, 2022 David Schor Arm Unveils Next-Gen Flagship Core: Cortex-X3 Mobile Processors Arm Unveils Next-Gen Flagship Core: Cortex-X3 June 28, 2022June 28, 2022 David Schor Arm Introduces The Cortex-A715 Architectures Mobile Processors Arm Introduces The Cortex-A715 June 28, 2022June 29, 2022 David Schor Alibaba Open Source XuanTie RISC-V Cores, Introduces In-House Armv9 Server Chip Architectures Server Processors Alibaba Open Source XuanTie RISC-V Cores, Introduces In-House Armv9 Server Chip October 20, 2021October 20, 2021 David Schor Marvell Launches 5nm Octeon 10 DPUs with Neoverse N2 cores, AI Acceleration Data Processing Unit Marvell Launches 5nm Octeon 10 DPUs with Neoverse N2 cores, AI Acceleration June 28, 2021June 28, 2021 David Schor Arm Introduces Its Confidential Compute Architecture Architectures Arm Introduces Its Confidential Compute Architecture June 23, 2021June 23, 2021 David Schor About WikiChip WikiChip is an independent publisher based in New York. The WikiChip Fuse section publishes chips and semiconductor related news with our main site offering in-depth semiconductor resources and analysis. WikiChip Links * Main Site * WikiChip Fuse * Newsletter * * Main Site * WikiChip Fuse Copyright (c) 2022 WikiChip LLC. All rights reserved. Spelling error report The following text will be sent to our editors: Your comment (optional): [ ] [ ] [ ] Send Cancel [noscript-1]