SRAM vs logic scaling with mfg processes - temporary situation, not a real trend

By: Heikki Kultala (heikk.i.kultal.a.delete@this.gmail.com), March 28, 2022 11:15 pm
Room: Moderated Discussions
I'm not 100% sure about this, but to me, the situation looks like this:

Lately, SRAM density has been scaling worse than logic density with new mfg processes, but it seems that this is not really a true trend, but just temporary situation:

Previously, SRAM has been like >5x more dense than logic (number of transistors/area). Because SRAMs have been consisting of highly density-optimized arrays where transistors have been packed very tightly.

Now, the scaling of manufacturing processes has slowed down and we have the "fake marketing nanometers" that have no correlation with real physical nanometers.

New manufacturing process has 1.4x smaller marketing number, but in reality has only something like 1.2 x smaller dimensions. And SRAM scales as it should, becoming ~1.44x denser with the new process.

But at the same time, the high density standard cells are being optimized more. Because of these optimizations, logic synthesized using the most high-density cells for the new mfg process has higher density improvements.

And because of the higher density standard cell logic density is getting better, the foundries dare to lie more about their "nanometers". The dare to call it with 1.4x smaller number even when the dimensions have only scaled about 1.2x.

And now the SRAM is getting to be only something like 3x denser (transistors/area) than logic, and may come down to maybe being only 2x denser. However, SRAM will always have higher transistor density than logic because they highly regular SRAM structures will be able to be packed more tightly. Logic is only narrowing the gap, never reschign the SRAM density.


So, the current "SRAM scaling worse than logic" seems to be just a temporary situation, due to logic density optimizations narrowing the gap to SRAM ensity optimizations.

And when making designs that go to future, we should not base them on a assumption that relatively to logic, "SRAM will keep getting more expensive".
 Next Post in Thread >
TopicPosted ByDate
SRAM vs logic scaling with mfg processes - temporary situation, not a real trendHeikki Kultala2022/03/28 11:15 PM
  SRAM vs logic scaling with mfg processes - temporary situation, not a real trend---2022/03/29 09:24 AM
    SRAM vs logic scaling with mfg processes - temporary situation, not a real trendDoug S2022/03/29 10:10 AM
    SRAM vs logic scaling with mfg processes - temporary situation, not a real trendme2022/03/29 06:06 PM
    SRAM vs logic scaling with mfg processes - temporary situation, not a real trendtarlinian2022/03/29 06:56 PM
      SRAM vs logic scaling with mfg processes - temporary situation, not a real trend---2022/03/29 07:54 PM
        SRAM vs logic scaling with mfg processes - temporary situation, not a real trendDoug S2022/03/29 10:30 PM
          SRAM vs logic scaling with mfg processes - temporary situation, not a real trend---2022/03/30 12:04 PM
            SRAM vs logic scaling with mfg processes - temporary situation, not a real trendDoug S2022/03/30 01:19 PM
      SRAM scaling and 3D "register file" design?Paul A. Clayton2022/04/25 09:55 AM
        SRAM scaling and 3D "register file" design?---2022/04/25 02:29 PM
          SRAM scaling and 3D "register file" design?Paul A. Clayton2022/04/26 11:40 AM
            SRAM scaling and 3D "register file" design?---2022/04/26 04:31 PM
  SRAM scaling problemsDavid Kanter2022/04/01 02:02 PM
  IMEC article about nanowire transistorsDoug S2022/04/05 12:08 PM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell avocado?