Russia sells S-400 systems to China

Anlsvrthng said:

(remarwk - I am not expert on this field, this is the informations that I collected to understand why the cost decrease of ICs stopped years ago)
The semiconductor industry using 193nm lasers even for 10nm parts.
The only way to do that is using multiple patterning, it means that the final pattern generated in several step, using interference patters from different masks.

However, the cost composition is a bit deeper, because two third of the semi plants come from cleaning/air filtration, the leftover from actual machines.

Problem is multiple patterning (And many other issue with smaller features) increase dramatically the cost of IC design.
IF a country learning the IC design, and has to make a lot of different type the simplest is to use 90nm technology, that gives the cheapest design for low volume ICs.

Click to expand...

gelgoog said:

There is more to it than that. The industry moved to so called immersion lithography. i.e. the chips are immersed into a liquid during the lithography process which enables etching finer details. It is a bit more complicated to use but it was cheap enough that it basically lead to the demise of the efforts to design lower wavelength light sources for like a decade.

Please, Log in or Register to view URLs content!

Unfortunately even with immersion lithography you can only go so far. That's why a lot of people claim that 45 nm is the manufacturing node which makes the most sense in the perspective of cost per transistor. To go beyond that you typically need to rely on multiple patterning which reduces the wafer output like you said. Multiple patterning allows you to etch finer details at the cost of more exposure steps. Not only does it cut wafer output but you also need to spend more money on photolithography masks. Which are typically one of the major costs in moving a new chip from design to manufacturability.

Click to expand...

Anlsvrthng said:

Under 90 nm either there is more mask, or Immersion, under 64nm both required.
However even the immersion require additional coating steps, so from the perspective of someone learn the trade the 90nm is the cheapest point.

Click to expand...

gelgoog said:

Right, the costs increase with each manufacturing node. This is called Rock's Law or Moore's Second Law.

Please, Log in or Register to view URLs content!

The only question is if the costs per transistor increase, decrease, or remain the same. Russia has more issues vs China in that they have a smaller chip market. Their population is much lower than that of China so there are less consumers too. So making a large chip factory makes less sense given their market conditions. Also AFAIK beyond the light source issues there is the fact that the industry has moved from 200 mm wafers to larger 300 mm wafers around the 45 nm node. This was done to reduce the cost per transistor but it means that manufacturing of smaller batches is more expensive as is wafer cost. Fabs like TSMC make do by placing multiple chip designs, from different manufacturers, on the same wafers so they can waste as little wafer die area as possible. But this means you have to make the customers use the same design rules and you have a harder time to get enough production volume to justify the operational costs.

Click to expand...

Anlsvrthng said:

Agree with it, but in case of Russia / China we talk not about the whole chip segment, but only a small, specialised portion of it, the military /radhard ICs.

Even the USA has hard time to keep this market segment alive, due to the low production volume.

Now, Russia problem is they need to develop ICs in the production range of dozens for satellites, in thousands for missiles, maybe ten thousands for other applications.
The only that maybe needs millions is the government systems.

Now, the USA has military IC designs, validated and production ready for very long time.

But now Russia has to make thousands of IC designs.

This means there is only limited amount of engineering resources.

Multi patterning, immersion lithography and so on makes more complicated the chip design, and longer the validation process, require more time and resources to do.

So, it is easier to make everything on 90nm, the part that makes big difference and can not make parallel can be done on smaller at same point of future.

Example the radar computers are very heavy parallel types, doesn't really matter if there is one or ten cpu doing the same job.

Click to expand...

gelgoog said:

For military ICs they would be interested in using SOI or FD-SOI manufacture and using different materials. i.e. other than silicon. SOI is inherently somewhat rad-hard but it requires the use of different design tools and techniques compared to more conventional manufacturing methods. For the case of radar chips the technology is not that different from telecommunications systems chips. So, say, China already has largely made their own solutions in this area. The Russians also have their own Elbrus 2K architecture which should have great performance for digital signal processing. The Chinese have their supercomputer architectures which would enable them to do much the same thing. Simpler systems can use the Elbrus chips with the SPARC architecture. SPARC, much like x86, also has vector instructions. So I wouldn't be surprised if the Elbrus SPARC chips also had those to facilitate digital signal processing. Yes their requirements are much more limited. That's why it's hard for Russia to justify the investment in fabs much better than the ones they have right now. Even if they were hit with an USA led embargo, which meant they could not use western factories to manufacture their chips, the Russians could still use Chinese fabs like SMIC for the more intricate chip designs. They have quite limited requirements in terms of chip production capacity. Still I think Russia has severe issues with their chip packaging technology, their integrated memory technology, and their production processes are utterly obsolete. They should have at least one 28 nm fab I think.

Click to expand...

gelgoog said:

When you mention IR sensors are you talking about thermal gun sights? Russia start by importing thermal gun sights for their tanks from France. Initially they bought the whole thing from France. Thales manufactured the sights. Then they switched to the Catherine-FC system which mostly used Russians manufactured parts except for the IR sensor which was still imported from France. Later they made their own system. This article describes the history rather accurately.

Please, Log in or Register to view URLs content!

Click to expand...

Anlsvrthng said:

Yes, this was the article that I read.

According this they had to develop a lot of new electronics ( including ICs) for the Russian supply chain.

Click to expand...

gelgoog said:

Right, the costs increase with each manufacturing node. This is called Rock's Law or Moore's Second Law.

Please, Log in or Register to view URLs content!

The only question is if the costs per transistor increase, decrease, or remain the same. Russia has more issues vs China in that they have a smaller chip market. Their population is much lower than that of China so there are less consumers too. So making a large chip factory makes less sense given their market conditions. Also AFAIK beyond the light source issues there is the fact that the industry has moved from 200 mm wafers to larger 300 mm wafers around the 45 nm node. This was done to reduce the cost per transistor but it means that manufacturing of smaller batches is more expensive as is wafer cost. Fabs like TSMC make do by placing multiple chip designs, from different manufacturers, on the same wafers so they can waste as little wafer die area as possible. But this means you have to make the customers use the same design rules and you have a harder time to get enough production volume to justify the operational costs.

Click to expand...