Chinese semiconductor thread II

[european_guy]

N+2 yield is close to 40%

https://twitter.com/i/web/status/2003070478987198519

Please take this with a grain of salt. Anyhow here is the translated page.

I'm a bit skeptical on the current capacity for N+2 set at 5K wpm (too low). He says that it will triple to 15K wpm next year, so now it is at 5K wpm. I was expecting 10-15K wpm _already_ now.

 

[latenlazy]

N+2 yield is close to 40%

https://twitter.com/i/web/status/2003070478987198519

Please take this with a grain of salt. Anyhow here is the translated page.

I'm a bit skeptical on the current capacity for N+2 set at 5K wpm (too low). He says that it will triple to 15K wpm next year, so now it is at 5K wpm. I was expecting 10-15K wpm _already_ now.


View attachment 166641

Endless telephone games with a teenage kid lol.

 

[european_guy]

Endless telephone games with a teenage kid lol.

I don't see the added value of discrediting a post content just because it's author is "teenager"...maybe I'm wrong.

I'm skeptical too but because of the content that seems at odd against various "hints" we got in the last months. But because we got just hints and no official info, cannot be "back or white" evaluating a post.

 

[latenlazy]

I don't see the added value of discrediting a post content just because it's author is "teenager"...maybe I'm wrong.

The added value is whether a source has the experience to make good judgments about the information passed to them. Anyways you can do some back of envelope math to see if those yield and wfm figures line up with annual unit sales of phones and servers. ((Die area)/(Wafer area))*(wfm*yield*12). Don’t think the math checks out.

 

[sunnymaxi]

N+2 yield is close to 40%

https://twitter.com/i/web/status/2003070478987198519

Please take this with a grain of salt. Anyhow here is the translated page.

I'm a bit skeptical on the current capacity for N+2 set at 5K wpm (too low). He says that it will triple to 15K wpm next year, so now it is at 5K wpm. I was expecting 10-15K wpm _already_ now.


View attachment 166641

N+2 yield is way more than 40 percent.

Please, Log in or Register to view URLs content!

Huawei even started to use N+2 on their Non-flagship products. this also shows the production capacity of N+2.

Huawei Nova 15 series Ultra will be using Kirin 9010S SoC and have 18% performance improvement vs previous generation.

just on last page, tphuang posted this link.

 

[AETHER]

N+2 40% yield may refer to 910C size die

 

[latenlazy]

N+2 40% yield may refer to 910C size die

If yield is 40% *AND* WFM is only 5,000 it’s very unlikely Huawei would have enough chips for their whole product line.

 

[sunnymaxi]

If yield is 40% *AND* WFM is only 5,000 it’s very unlikely Huawei would have enough chips for their whole product line.

yup. Huawei literally swimming in chips. they even started to use N+2 in Non-flagship Nova series.

Please, Log in or Register to view URLs content!

Huawei Nova 15 series Ultra will be using Kirin 9010S SoC and have 18% performance improvement vs previous generation.

 

[zzz]

N+2 40% yield may refer to 910C size die

Yes. Context is important:

 

[tphuang]

N+2 yield is not a constant thing. it's clearly going to be different for different chips (depending on size and chip complexity) and binning and such. Ascend chips are a lot larger, but it also has lower complexity. Hard for someone from outside like us to assess yield on K8000 vs 910B die for example.

The production level and yield are not unreasonable if we assume that's just for AI chips. 40% for a 400-500mm2 AI chip and for 14-15k wpm. okay. Well, they've had 20-30k wpm overall for a couple of years. So if next year, they expand to 60k, then dedicating 15k to AI chips and some for server CPUs and 30k for Huawei SoCs, probably reasonable.

But it's not clear looking at that.

I found the part about using 12nm process to process a lower spec'd PPU to be not so convincing. I mean, what are you going to do with that? You need at least N+2 process, I would imagine.

But that line of reporting seems kind of strange.

fyi, expect 950 to be smaller die than 910B.