Node size (7nm, 5nm, etc) is now pointless

[NotFuzzyWarm]

As insiders have known for many years now, a chips' node size has become a pointless and erroneous measure of what the dimensions inside of a chip actually are and not-so-useful in defining what the chip performance can be.

Looks like a new definition is in the works to better reflect actual progress being made. ref https://read.nxtbook.com/ieee/spectrum/spectrum_na_august_2020/the_node_is_nonsense.html

My take on it is that companies like MicroBt and Canaan made the right choice by staying at 16nm for so long and relying on the fact that the Market forces driving the quest for lower 'node size' for things like memory, CPU's, GPU's etc as well as mining chips are at the same time benefiting the higher nodes. How?

Easy - use for example the '16nm node' that MicroBt and Canaan used for their respective M10/M20 and A10xx series miners. My guess is that TSMC applied lower-node metal interconnect tech to the now much cheaper 16nm node. Smaller/thinner metalization interconnect layers means lower switching losses improving efficiency while still using the larger (and cheaper, fully mature) 16nm gate size. It keeps the long ago paid for 16nm foundries busy while providing decent performance chips at far lower cost than cutting edge chips like BM's 7nm and now their bleeding-edge 5nm node size...

AFAIK the M30x from MicroBT use what, 12nm 8nm chips? I know Canaan is still smarting from the 7nm A921 debacle: They and/or Samsung rather blew it on that chip so even now their latest A1066 is still a brick of 16nm chips sourced from TSMC.

Edit: Earlier this year Canaan mentioned an A1166 but not sure what 'node size' it is/was supposed to be. They've been mighty quiet about it ever since.
Edit edit: The A11's specs say it has 342 of their A3205 16nm ASIC

[Biffa]

AFAIK the M30x from MicroBT use what, 12nm chips?

8nm Samsung

[HagssFIN]

It's quite impressing what Canaan has been able to reach with 16nm tech and their A1166 Pro miner.

[NotFuzzyWarm]

Ja, both Canaan and MicroBt have really pushed 16nm node (transistor 'gate size') performance to its probable maximum. As I said, my guess is that they did it by using smaller/thinner metal layers made economically possible by the advances needed for the lower nodes.

[Biffa]

Anyone know how much the 1166 Pro is? Its not for sale on Blokforge.

[DaveF]

Is now pointless is a bit extreme. Is pointless at the moment is probably better. You will get performance increases and you will get energy savings. IF it's done properly. There are other ways to get performance and energy savings but there are limits. Once you hit those limits you will have to go smaller / more efficient. OR go another way entirely.

Think of the car world, gas engines have gotten more and more efficient. Then came hybrids / electric. There might be a better way lets say hydrogen fuel cell, but there are other issues such as filling stations. If you go for same analogy. Canaan and MicroBt have gone for more efficient gas engines. The smaller node size is electric. We don't know who has hydrogen waiting to be shown.

-Dave

[philipma1957]

Yeah and a gas engine for a car won't do better than 50 mpg.

They may have reached the best a 16nm can do.

Down the road a bit pow mining of sha 256 will have other issues. Even with a 300th 3000watt unit When block size is 0.78125 or 1.5625 coins we will need proper tx fee setup and or 100k coins.

[NotFuzzyWarm]

... so the point doesn't matter what size node is, the only matter is efficiency.

Yes and No. Yes, what matters is the efficiency of any given "node size", that in turn strongly depends on how good the chip designers are at laying out the circuits inside of the chip.

No, the point is that for many years now the term "node size" says nothing about ANY specific physical dimensions inside of a chip. It is now just a number that Marketing uses to say one manufacturing technology (node) has gate features that are smaller or larger than another technology (node). Problem is, gate size is just 1 part of the design and shrinking that (more) is increasingly becoming less important compared to say design of the metal layers.

When done correctly smaller does equal less power for same speed or more speed for same power. The 'gotcha' comes from 'done correctly'. There are many many ways to screw up chip design. As said in opening post - Canaan proved that with their then bleeding-edge 7nm A921 miner whose efficiency was nearly the same as the A8xx series that have 3rd generation 16nm chips.

Anyone know how much the 1166 Pro is? Its not for sale on Blokforge.

Well sorta maybe is? https://blokforge.com/product/canaan-avalon-1166-68-th-s-bitcoin-miner-w-psu-february-2020-batch/. Says it still uses 16nm chips

[Biffa]

Well sorta maybe is? https://blokforge.com/product/canaan-avalon-1166-68-th-s-bitcoin-miner-w-psu-february-2020-batch/. Says it still uses 16nm chips

Nah, thats just the normal (non-pro) 1166 - 68TH @ 47J/TH or this: https://canaan.io/product/avalonminera119

The 1166-Pro - 81TH @ 42J/TH or this: https://canaan.io/product/avalonminer-1166-pro isn't showing on the Blokforge site. And of course Canaan are only interested in bulk orders  

https://canaan.io/products/avalonminer

[NotFuzzyWarm]

Further reading to prove my point about using just "node size" now being a pointless measure of what a chip's performance can be and why the Semiconductor industry working on a better descriptor:

On replacing Si with Ge as the semiconductor

On using entirely different principal of switching transistor operation

On using an entirely different type of transistor geometry vs current FinFET's

The 1st 2 papers are not exactly recent but still very valid examples of what is being worked on. The last paper is from last year.
All are very good reads that give a lot of insight into how chips operate and are made along with the difficulties that have been mitigated to be where we are currently at and those that have yet to be overcome to advance further/faster.