Chinese semiconductor industry

Huawei 5G radios: 12 nm & chiplets should do the job​

Although the best mobile phones need the most advanced chips to be competitive, 5G radios simply have to be able to perform their functions effectively. Radios, unlike phones, have room and power for 2 or 3 slightly less efficient chips. The necessary functions can be split between two chips or three chips, each optimized for the many different functions required by the 5G radio. Huawei will neither confirm nor deny this is practical, so I am reaching out to chip experts.

There are many different ways to measure efficiency and experts disagree on how much more efficient 7 nm chips are than 12 nm, the best that can be produced today at SMIC in China. Estimates range from 20% to 50%, and there’s no simple way for a firm estimate until chips are available for test. Almost certainly, two 12 nm chips can match the functions of one 7 nm chip. A third chip could be optimized for functions that needed the most speed.

The 2 or 3 chips can be delivered as “chiplets,” which AMD and others are demonstrating can be very effective. AMD’s Mark Papermaster is the pioneer here, using a chiplet design to quickly bring the Epyc server chip to market.

“I think the whole industry is going to be moving in this direction,” Smaller chips are much faster and less expensive to design. They also are less sensitive to defects than a monolithic chip. On a monolithic chip, a single defect means it must be destroyed. If broken into three separate chips, the other two parts are still fine. AMD engineers estimate the cost is halved.

Ramune Nagisetty of Intel agrees. She calls it “an evolution of Moore’s law. Intel has a very deep roadmap for chiplets. This is the future.” Intel is already shipping a processor for mobile PCs that combines an Intel CPU with a graphics module from AMD.

Of course the performance of 12 nm chips does not match what’s possible in 7 nm. But with good design, much of the gap can be closed. There’s a cost involved, but it’s not material to a radio selling for $5-10,000.
Tough years ahead if the U.S. continues the blockade but Huawei will survive.

krautmeister said:

Please, Log in or Register to view URLs content!

Huawei 5G radios: 12 nm & chiplets should do the job​

Although the best mobile phones need the most advanced chips to be competitive, 5G radios simply have to be able to perform their functions effectively. Radios, unlike phones, have room and power for 2 or 3 slightly less efficient chips. The necessary functions can be split between two chips or three chips, each optimized for the many different functions required by the 5G radio. Huawei will neither confirm nor deny this is practical, so I am reaching out to chip experts.

There are many different ways to measure efficiency and experts disagree on how much more efficient 7 nm chips are than 12 nm, the best that can be produced today at SMIC in China. Estimates range from 20% to 50%, and there’s no simple way for a firm estimate until chips are available for test. Almost certainly, two 12 nm chips can match the functions of one 7 nm chip. A third chip could be optimized for functions that needed the most speed.

The 2 or 3 chips can be delivered as “chiplets,” which AMD and others are demonstrating can be very effective.

Please, Log in or Register to view URLs content!

AMD’s Mark Papermaster is the pioneer here, using a chiplet design to quickly bring the Epyc server chip to market.

“I think the whole industry is going to be moving in this direction,”

Please, Log in or Register to view URLs content!

Smaller chips are much faster and less expensive to design. They also are less sensitive to defects than a monolithic chip. On a monolithic chip, a single defect means it must be destroyed. If broken into three separate chips, the other two parts are still fine. AMD engineers estimate the cost is halved.

Ramune Nagisetty of Intel agrees. She calls it “an evolution of Moore’s law. Intel has a very deep roadmap for chiplets. This is the future.” Intel is already shipping a processor for mobile PCs that combines an Intel CPU with a graphics module from AMD.

Please, Log in or Register to view URLs content!

Of course the performance of 12 nm chips does not match what’s possible in 7 nm. But with good design, much of the gap can be closed. There’s a cost involved, but it’s not material to a radio selling for $5-10,000.
Tough years ahead if the U.S. continues the blockade but Huawei will survive.

Click to expand...

@krautmeister bro again a nuisance question hope you can bear with me, is this the panacea that Huawei is looking for especially if it can be done at 7nm node level, a 7NM 3D chiplet comparable to TSMC 5nm? cause as @WTAN had posted 2025 is the year that a likelihood of an EUVL introduction, so it will be a transitory chip before moving to a 5nm or 3nm chip?

Here's an interesting snippet about Intel's 7nm CPU chiplet plans. Variations of this chiplet CPU will be geared towards different user segments like productivity, gaming, content creation, mobile, etc. Supposedly, Intel is having big headaches with their chiplet integration and packaging now delayed to 2022, possibly 2023.


In this diagram, we are treated to Intel’s long term vision for the client – a base interposer with an in-package memory (something like an L3 or L4) that can act as the main SRAM cache for the whole die, and then on top of this we get 24 different chiplets. Chiplets can be graphics, cores, AI, Media, IO, or anything else, but they can be mixed and matched based on what is needed. A content creator might want a balance between some good graphics acceleration and compute, while a gamer might want to focus purely on the graphics. A corporate client or workstation might need less graphics and more for compute and AI, whereas a mobile version of the chip will be heavily invested in IO. As always, there is some trade-off between chiplet size and complexity of actually putting them together in a multi-die arrangement.

ansy1968 said:

@krautmeister bro again a nuisance question hope you can bear with me, is this the panacea that Huawei is looking for especially if it can be done at 7nm node level, a 7NM 3D chiplet comparable to TSMC 5nm? cause as @WTAN had posted 2025 is the year that a likelihood of an EUVL introduction, so it will be a transitory chip before moving to a 5nm or 3nm chip?

Click to expand...

It depends on what you include as the panacea. I posted something about Intel's 7nm chiplet plans for their CPUs and they have been confronted with multiple delays, primarily related to packaging. However, packaging is 1 of China's strengths, but it's starting late and assuming it does have successful 7nm trials, by then the goalpost for the leading edge will have moved to 3nm EUVL, possibly even 2nm. This is actually not that important if we are not considering smartphones because 7nm is already pretty advanced and it covers well over 90%+ of what China needs. That includes AI, IoT, CPUs, GPUs, controllers, etc, etc, etc. These markets would open up to China at the leading edge of those product segments. This is what we should be considering when we talk of the panacea.

Either way, as @BoraTas suggested, this is a very good option to keep Huawei mobile division competitive until they have EUVL. Not that China even has a choice, besides doing a Japan. There are ways to optimize the chiplet design to improve performance but the power draw will be greater. It won't have leading edge performance but it wouldn't be a slouch either. Right now, the Harmony OS most talked about is the Android fork+HMS that is most often tested on smartphones. The other version of Harmony OS is the kernel version that is installed on all other devices. It is this version that I believe will be installed in future versions of Huawei's phones which are the Harmony kernel+AOSP+HMS installation. With this combination, Huawei would achieve leading edge performance for native apps. Will this succeed? In China, very likely because China doesn't rely on GMS (Google Services) anyways and has its own fully developed app ecosystem. Outside China, very very tough outside the enthusiast tech community.

Realistically, this is what we should expect for now until domestic EUVL is available. Until then, forget about Huawei's mobile market outside China. That will depend on foreign processors whose supply I expect to be cut off at some point in the future.

Tsinghua has some problems.

@WTAN
Do you know more?

krautmeister said:

It depends on what you include as the panacea. I posted something about Intel's 7nm chiplet plans for their CPUs and they have been confronted with multiple delays, primarily related to packaging. However, packaging is 1 of China's strengths, but it's starting late and assuming it does have successful 7nm trials, by then the goalpost for the leading edge will have moved to 3nm EUVL, possibly even 2nm. This is actually not that important if we are not considering smartphones because 7nm is already pretty advanced and it covers well over 90%+ of what China needs. That includes AI, IoT, CPUs, GPUs, controllers, etc, etc, etc. These markets would open up to China at the leading edge of those product segments. This is what we should be considering when we talk of the panacea.

Either way, as @BoraTas suggested, this is a very good option to keep Huawei mobile division competitive until they have EUVL. Not that China even has a choice, besides doing a Japan. There are ways to optimize the chiplet design to improve performance but the power draw will be greater. It won't have leading edge performance but it wouldn't be a slouch either. Right now, the Harmony OS most talked about is the Android fork+HMS that is most often tested on smartphones. The other version of Harmony OS is the kernel version that is installed on all other devices. It is this version that I believe will be installed in future versions of Huawei's phones which are the Harmony kernel+AOSP+HMS installation. With this combination, Huawei would achieve leading edge performance for native apps. Will this succeed? In China, very likely because China doesn't rely on GMS (Google Services) anyways and has its own fully developed app ecosystem. Outside China, very very tough outside the enthusiast tech community.

Realistically, this is what we should expect for now until domestic EUVL is available. Until then, forget about Huawei's mobile market outside China. That will depend on foreign processors whose supply I expect to be cut off at some point in the future.

Click to expand...

@krautmeister bro again thanks for your time and patience, learned a lot.

Apologies if this has been posted. But I just came across this. Why would China want to buy old tech company in the U.K.? Apperently there Is still a market for this.

Semiconductors: Chinese takeover of UK's leading chipmaker doesn't need a security review – here's why

Wingtech Technology is taking over Newport Wafer Fab for £65 million.

Link:

krautmeister said:

However, packaging is 1 of China's strengths, but it's starting late and assuming it does have successful 7nm trials, by then the goalpost for the leading edge will have moved to 3nm EUVL, possibly even 2nm. This is actually not that important if we are not considering smartphones because 7nm is already pretty advanced and it covers well over 90%+ of what China needs.

Click to expand...

I'm still a little doubful about the result of the sub 5nm race. Whether one can make 2-3nm chips is one thing, whether those chips are any good is another. Looking at 5nm Exynos by Samsung foundry having problems. So many Korean users of Samsung Galaxy S21 series are complaining their phones getting unsusably/unbearably hot and apps get throttled/force-closed due to overheating, and they can't do even basic tasks. Some say Samsung 5nm is comparable to TSMC 7nm. I guess much part of 3nm and 2nm media coverage are still advertizing and marketing hypes by the foundry to show off their technological supremacy. TSMC might be different though.

Gatekeeper said:

Apologies if this has been posted. But I just came across this. Why would China want to buy old tech company in the U.K.? Apperently there Is still a market for this.

Semiconductors: Chinese takeover of UK's leading chipmaker doesn't need a security review – here's why

Wingtech Technology is taking over Newport Wafer Fab for £65 million.

Link:

Please, Log in or Register to view URLs content!

Click to expand...

There have been several comments about that subject on page 646, 647 after this post:

KYli said:

Not sure if the UK would block China from buying a low end chip plant.

Please, Log in or Register to view URLs content!

Click to expand...

ILikeChina said:

Please, Log in or Register to view URLs content!

Tsinghua has some problems.

@WTAN
Do you know more?

Click to expand...

Sounds like China wants to kick some problematic shareholders out via a restructuring.