Chinese semiconductor industry

european_guy said:

In the article is also written "The results were achieved while using a relatively old process node of 55 nm for the logic"

If it is confirmed, then this is a real breakthrough, and to me it is the most important point of this article. It means state-of-the-art performance in AI could be achieved with semiconductor equipment already (almost) localized in China.

Click to expand...

What would happen if it were with newer process node of 14nm and 7nm and with Chinese design? Would the performance become 3x or 5x better ?

This is a similar AI chip from a US company (Graphcore).

It uses TSMC 7nm process. The chip is made from two wafers bonded together with TSV.

I think this analog AI chip from another US company (Aspinity) is more viable for older process.

Analog AI approach needs a lot less power and transistors.

The "Collective West" has placed disastrous sanctions on Venezuela and Iran, now on Russia.

The US now is desperate to finalize the JCPOA deal in order to boost Iranian sales of oil and has sent a delegation to Venezuela in order to get oil from Venezuela. Can you imagine to what point their desperation has come?

Just heard from a friend of mine in the Venezuelan embassy that one of the conditions of supplying oil now is that they be paid either in Petroyuans or gold.

It seems to me also that Russia and Iran may join the demand.

Can you imagine the US having to buy oil paying in Petroyuan!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

Due to the upcoming war in all of Europe, China is writing Europe off.

• Chinese premier puts the focus on secure industrial chains and technological breakthroughs

gelgoog said:

This is a similar AI chip from a US company (Graphcore).

Please, Log in or Register to view URLs content!

It uses TSMC 7nm process. The chip is made from two wafers bonded together with TSV.

I think this analog AI chip from another US company (Aspinity) is more viable for older process.

Please, Log in or Register to view URLs content!

Analog AI approach needs a lot less power and transistors.

Click to expand...

I think Graphcore is from the UK.
The problem with analog computing for AI is the same as the hype for memristors for AI if your average Joe can't program it then i don't think will go anywhere.

tokenanalyst said:

I think Graphcore is from the UK.
The problem with analog computing for AI is the same as the hype for memristors for AI if your average Joe can't program it then i don't think will go anywhere.

Click to expand...

As long as it can process Tensorflow/Keras then it's good.

jfcarli said:

The "Collective West" has placed disastrous sanctions on Venezuela and Iran, now on Russia.

The US now is desperate to finalize the JCPOA deal in order to boost Iranian sales of oil and has sent a delegation to Venezuela in order to get oil from Venezuela. Can you imagine to what point their desperation has come?

Just heard from a friend of mine in the Venezuelan embassy that one of the conditions of supplying oil now is that they be paid either in Petroyuans or gold.

It seems to me also that Russia and Iran may join the demand.

Can you imagine the US having to buy oil paying in Petroyuan!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

Click to expand...

dfrtyhgj said:

Due to the upcoming war in all of Europe, China is writing Europe off.

• Chinese premier puts the focus on secure industrial chains and technological breakthroughs

Please, Log in or Register to view URLs content!

Click to expand...

I think this should be posted in the China Economic thread.

caudaceus said:

As long as it can process Tensorflow/Keras then it's good.

Click to expand...

Not exactly. Analog AI needs whole new software. The world is so fucking deep into the Von Neumann architecture that I doubt it will ever be replaced.

antiterror13 said:

What would happen if it were with newer process node of 14nm and 7nm and with Chinese design? Would the performance become 3x or 5x better ?

Click to expand...

when you get to that level you have quite a bit of dark silicon (unused transistors) due to thermal overload if all transistors were used at once, which is why mere die shrinks are hitting the wall. remember back 20 years ago that clock speeding up was the way to add speed but then it stopped working, then it was die shrinks to put multiple cores on 1 die but now it stopped working too, then it was SoC to put peripherals on the same die as well.

in terms of process you had to move from planar transistor to finFET for further die shrinks, you had to use metal gate instead of polySi gate, you had to use high-k instead of silicon dioxide insulator, etc. Beyond just lithography, the entire chemical processing of the chip changed. it's a paradigm shift.

now you're at the limit, hitting shrinking return on investment, and need a new paradigm. the 3D packaging integration is the new paradigm. so mere die shrink in this case may or may not increase performance but it won't be 3x-5x.

SMEE 2021 patent for their DUVi machine.

Installation and adjustment device of grating ruler on immersion head​

Abstract​

The utility model provides an installation and adjustment device for a grating ruler on an immersion head. The utility model provides an installation and adjustment device for a grating ruler on an immersion head. For a plurality of grating rulers fixed on the immersion head included in the immersion lithography machine, the installation and adjustment device provided by the utility model is installed on each grating. The attachment of the ruler is provided with a height-adjusting component for adjusting the movement of the immersion head along the preset six-degree-of-freedom direction, and a positioning component for adjusting the positional relationship between the grating ruler and its corresponding reading head. The height-adjusting component and the positioning component can realize the integrated adjustment of multiple grating scales, thereby reducing the adjustment difficulty of the grating scales and improving the measurement accuracy of the immersion head pose measurement system.