Chinese semiconductor industry

Status
Not open for further replies.

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
We really need to calm down a little bit. Progress is incremental, no revolutionary. I would also say the same here

Took time for SMIC to develop N+1, probably using NXT1980i. After it got NXT 2050i, it was able to do N+2. And it also helped to get the better etching equipment from Lam/AMEC that @latenlazy was referring to.

If we regard TSMC N5 as the golden standard for 5nm (which I think is what is considered by Tech Insight), then I don't N+2 is there based on everything we read, but it's pretty far advanced in transistor density and power consumption (i mean it's power consumption for A510 cores are better than Samsung 4nm process). That's pretty insane.

I also think @latenlazy makes a good point that Huawei has time to make up edge with Snapdragon with Kirin 9100, since Qualcomm is going without N3 process with SD 8 Gen 3 and maybe 4 (depends on apple here)

And there are plenty of improvement left for Hisilicon here.

The important progress outside of what SMIC has done include:
BAW filter & L-PAMiD by Hisilicon so that it can make it's own 5G (or 5.5G as HW fanboys call it) RFFE
Satcom RF & processing chip integration with rest of RF
the amazingly efficient heat dissipation tech
new Taishan core
new Maleeon GPU
upgraded Da Vinci NPU
new 5G modem

For Kirin 9100, I think we need to see
SMIC N+2 improved with higher yield & transistor density. Something close to N5 at 75% yield would be nice
Greater cache size (L3 cache on 9000S is pitifully small)
Improved Taishan core (is hyper-threading necessary on a phone?)
Improved Maleeon GPU
Improved power consumption for both CPU & GPU (not just at lower load)
Improved NPU
More compact & efficient 5.5G modem
BAW filter that can fully handle 6GHz and higher band
Larger die size please (need it to get the cache and everything else in there)
Improved battery tech to deal with higher power consumption compared to latest Snapdragon 8

Whatever comes out needs to be competitive in performance with Snapdragon 8 Gen 3/4 while also not loose out completely in power consumption (at least at regular passive usage)

This covers them for 2024/2025

Beyond that, HW needs to work with partners to release new
SoC for wearables using 12nm process
SoC for screens using 12nm process
SoC for tablets/pads using N+2
Kunpeng replacement CPU for server to desktop to laptop
Ascend AI chip for replacing Ascend-910

More than anything else, they need to get new tools validated. So SMEs need to work a lot harder
The issue now is not necessarily tech itself, but rather capacity
What good is 25k wpm? They need 125k wpm of FinFet capacity

We are far away from finishing here
 

CMP

Senior Member
Registered Member
The biggest "shock" is probably achieving 5nm level capabilities with a 7nm node. It shouldn't be possible (if you ignore all the articles on advanced packaging and other alternate ways to squeeze out performance industry experts have talked about for the last 5 years)

What was meant to happen was the rapid sanctions would cause a general collapse of the industry with money pulling out, everyone resigning from SMIC and other sanctioned companies because they would be doooomed. Progress on catchup R&D stalls for years and years, turning a 10-year lead into 20-year lead. Hi-silicon shouldn't be able to maintain any sort of leading edge R&D at all for Huawei to remain competitive following that logic.

The BIS wonks obviously missed something, or talked to the wrong "experts" and created an echo chamber. Novel sanctions are called novel, for a reason.
Yup. It was firstly an economic war by means of tech war, following on the trailing edge of a trade, hacking, spying, propaganda, and diplomatic war. Literally every type of war is being waged by the American government against China, and preparations for actual war are in progress. It would be so satisfying to see reciprocal actions, but it seems China is responding only via its traditional method of acupuncture-style geopolitics. Extremely formulated and well-calibrated actions that are usually asymmetric and never excessive. In some ways it's highly unsatisfying for peanut gallery like us, but it's the right thing to do for the best path to victory.
 

vincent

Grumpy Old Man
Staff member
Moderator - World Affairs
What was meant to happen was the rapid sanctions would cause a general collapse of the industry with money pulling out, everyone resigning from SMIC and other sanctioned companies because they would be doooomed. Progress on catchup R&D stalls for years and years, turning a 10-year lead into 20-year lead. Hi-silicon shouldn't be able to maintain any sort of leading edge R&D at all for Huawei to remain competitive following that logic.

The BIS wonks obviously missed something, or talked to the wrong "experts" and created an echo chamber. Novel sanctions are called novel, for a reason.
BIS should have taken a look at the makeup of researchers in American research and development labs.
 

tinrobert

Junior Member
Registered Member
Those reports are not free. That's how they make money. Their subscribers are unlikely to post entire reports on public forums as it would be against their service agreement. But they would make some information public.
On May 18, 2022 I published an article that SMIC had reached the 7nm node, 2 months before Techinsights made its announcement.
You can read how they did it here, and it's FREE.
Please, Log in or Register to view URLs content!
 

Phead128

Captain
Staff member
Moderator - World Affairs
We really need to calm down a little bit. Progress is incremental, no revolutionary. I would also say the same here

Took time for SMIC to develop N+1, probably using NXT1980i. After it got NXT 2050i, it was able to do N+2. And it also helped to get the better etching equipment from Lam/AMEC that @latenlazy was referring to.

If we regard TSMC N5 as the golden standard for 5nm (which I think is what is considered by Tech Insight), then I don't N+2 is there based on everything we read, but it's pretty far advanced in transistor density and power consumption (i mean it's power consumption for A510 cores are better than Samsung 4nm process). That's pretty insane.

I also think @latenlazy makes a good point that Huawei has time to make up edge with Snapdragon with Kirin 9100, since Qualcomm is going without N3 process with SD 8 Gen 3 and maybe 4 (depends on apple here)

And there are plenty of improvement left for Hisilicon here.

The important progress outside of what SMIC has done include:
BAW filter & L-PAMiD by Hisilicon so that it can make it's own 5G (or 5.5G as HW fanboys call it) RFFE
Satcom RF & processing chip integration with rest of RF
the amazingly efficient heat dissipation tech
new Taishan core
new Maleeon GPU
upgraded Da Vinci NPU
new 5G modem

For Kirin 9100, I think we need to see
SMIC N+2 improved with higher yield & transistor density. Something close to N5 at 75% yield would be nice
Greater cache size (L3 cache on 9000S is pitifully small)
Improved Taishan core (is hyper-threading necessary on a phone?)
Improved Maleeon GPU
Improved power consumption for both CPU & GPU (not just at lower load)
Improved NPU
More compact & efficient 5.5G modem
BAW filter that can fully handle 6GHz and higher band
Larger die size please (need it to get the cache and everything else in there)
Improved battery tech to deal with higher power consumption compared to latest Snapdragon 8

Whatever comes out needs to be competitive in performance with Snapdragon 8 Gen 3/4 while also not loose out completely in power consumption (at least at regular passive usage)

This covers them for 2024/2025

Beyond that, HW needs to work with partners to release new
SoC for wearables using 12nm process
SoC for screens using 12nm process
SoC for tablets/pads using N+2
Kunpeng replacement CPU for server to desktop to laptop
Ascend AI chip for replacing Ascend-910

More than anything else, they need to get new tools validated. So SMEs need to work a lot harder
The issue now is not necessarily tech itself, but rather capacity
What good is 25k wpm? They need 125k wpm of FinFet capacity

We are far away from finishing here
So Kirin series has runway for iterative improvements to remain competitive with Snapdragons 8 gen 3/4 over next 1-2 years (2024-2025)!! What a beautiful insight. But by then, prototype EUV by CIOMP should be released in 2024 or 2025. To keep competitiveness, I wonder if advanced 2D stacking techniques can bridge this slight transition period before EUV enters serial production?

As you said, the tech is not really the issue but scaling the production to meet the demands. West believed China semiconductor is dead below 14nm. Now China has a capacity problem below 14nm, what a beautiful "problem" to have!!
 

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
Something to think about

Please, Log in or Register to view URLs content!

Sd8 gen 2 cost $160 each. Let's say 400 good Kirin 9000s per smic n+2 wafer.

If each smic wafer costs $17k (how much tsmc charges for n5)

Then, Huawei saves 400 × $160 - $17000 = $47000 per wafer which can go toward packaging/testing cost, hisilicon r&d and greater profit.

I generally see this work out economically for Huawei & smic.

Even if smic charges $20k per wafer, this seems profitable trade.
 

horse

Colonel
Registered Member
Yup. It was firstly an economic war by means of tech war, following on the trailing edge of a trade, hacking, spying, propaganda, and diplomatic war. Literally every type of war is being waged by the American government against China, and preparations for actual war are in progress. It would be so satisfying to see reciprocal actions, but it seems China is responding only via its traditional method of acupuncture-style geopolitics. Extremely formulated and well-calibrated actions that are usually asymmetric and never excessive. In some ways it's highly unsatisfying for peanut gallery like us, but it's the right thing to do for the best path to victory.

I had a thought about that, a related thought.

Why did Huawei did not even say what type of chip was inside the Mate 60 Pro and could it connect to a 5G network? People only confirmed it when the bought the phone, and reviewed the tear downs.

Why the secrecy, for an expensive product that is being sold?

Let's face it, it is not every day, we go out and buy an expensive product, and we don't know what it even does.

That seemed like the marketing campaign here. It was absolutely wonderful, because it delivered.

The point is this. This is still a tech war, and this was obviously an attack. In an attack, we always want the element of surprise.

This secrecy tells me, this is only the beginning of several waves of attack, and every which way too.

As of today, I believe Huawei, or anyone else, as neither confirmed or denied what is going on with the Mate 60 Pro.

Something more is being planned.

:oops:
 
Status
Not open for further replies.
Top