Chinese semiconductor thread II

gelgoog

Brigadier
Registered Member
Who knows. There were more companies in China which were supposed to make VNAND other than YMTC.

If you believe the information that has been getting out of China though the Kirin 9010 is pretty different from the 9000S unlike what this article claims. It uses a whole new Taishan large core design.

It would be interesting to check other Huawei smartphones to see if they use CXMT DRAM.
 

tonyget

Senior Member
Registered Member

Huawei’s Kirin 9010 Is a Reality Check for China’s Semiconductor Ambitions​


Please, Log in or Register to view URLs content!

Huawei released the Pura 70 line of smartphones in April 2024 and it’s already made a bit of a stir. Last year, another Huawei smartphone, the Mate 60,
Please, Log in or Register to view URLs content!
with an advanced 7nm-process System on Chip—despite US sanctions that limited Chinese access to the latest lithography machines. Sure, it was still years behind other semiconductor manufacturers, which had moved on to
Please, Log in or Register to view URLs content!
chips. But they caught up faster than most expected.

As the tech world anticipated this new smartphone, everyone wanted to know: Has China continued to advance its semiconductor manufacturing processes, in defiance of sanctions? Is the chip in the Pura 70 new? Better? Where is it made, and on what equipment?
To get some answers, we tore down the Pura 70 Pro in collaboration with our partners at
Please, Log in or Register to view URLs content!
.

A Closer Look at the Kirin 9010​

Thanks to those same sanctions, we couldn’t get a Pura 70 in our US office. But we got some help from Reuters, who worked with our team in Shenzhen to do a midnight teardown and get a closer look at the phone’s new silicon—which is, we discovered, the new Kirin 9010 System on Chip.

daniel-chip.jpg


Like many modern SoC packages, the Kirin 9010 processor sits underneath the DRAM module, the 12GB SK Hynix chip in this instance. Removing the chip reveals the processor.

Shenzhen-decap.jpg

The Kirin 9010 pictured from the teardown of a Pura 70 Pro

From this image, one detail stood out. The external markings seemed to match the older Kirin 9000S which was produced on SMIC’s 7nm node, commonly referred to as N+2. We ran this by one of our industry experts, and they noted that the chip identifiers in particular were unusual in that a new chip usually has its own model number. Not in this case: the 9000S (model HI36A0 revision GFCV120) and the 9010 (model HI36A0 revision GFCV121) share the same model number.
9000S-v-9010.jpg


In the days that followed, we saw similar sentiments appear online. TechInsights suggested that the Kirin 9010 appeared to be on the
Please, Log in or Register to view URLs content!
. The prevailing theory is that the Kirin 9010 is, at its heart, a 9000S with a revised design aimed at improving production yields. The improvements may not be limited to production though—early benchmarks suggest that the 9010 also performs a little better than the 9000S.

This is significant because news of the 9000S on a 7nm node
Please, Log in or Register to view URLs content!
last year when US lawmakers were confronted with the possibility that the sanctions imposed on Chinese chip makers
Please, Log in or Register to view URLs content!
their technological progress after all. The fact that the 9010 is still a 7nm process chip, and that it’s so close to the 9000S, might seem to suggest that Chinese chip manufacturing has indeed been slowed. But don’t underestimate Huawei. SMIC (Semiconductor Manufacturing International Corporation) is still expected to make the leap to an N5 equivalent node before the end of the year, if not shortly thereafter. They already possess the technology and know-how; it’s simply a matter of achieving commercially viable yields.
chip-ID.jpg

Still, there’s no question that the Chinese semiconductor industry is trailing several years behind TSMC, Global Foundries, Samsung, and other Western-aligned fabs with access to ASML’s EUV (Extreme Ultraviolet) photolithography machines. SMIC was unable to progress to the 5nm node process since the release of the 9000S while TSMC has an N2 node in its sights—a generational difference of 3 nodes representing six years of process refinement. That should put the technical challenges facing China’s semiconductor industry into perspective.

Looking at the benchmarks between the 9000S, 9010, and 2023’s Qualcomm Snapdragon 8 Gen 3 produced by TSMC shows how far Chinese semiconductor manufacturing has come, but also how far they have to go. The Kirin 9010 performs staggeringly worse than TSMC’s N4P derived Snapdragon 8 Gen3. N4P is a node process derived from the N5 node and announced by TSMC back in 2021 and its N5 parent node is only a single generation apart from the N7 node of the Kirin 9000S and 9010. Meanwhile, the performance increase between the 9000S and 9010 remains in single digits. This new SoC is marginally better but not the huge leap that was expected.

Benchmarks.jpg

Kirin 9000S (ALN-AL00) and Kirin 9010 (HBN-AL00) / Credit: Geekbench

To achieve those single-digit performance increases, the Kirin 9010’s 12-core SoC uses six ARM Cortex A510 efficiency cores running at 1550 MHz paired with four Taishan v121 cores running at 2180 MHz and two Taishan v121 cores running at 2300 MHz.
By comparison, the 9000S is reported to have an 8-core SoC (confusingly there are lots of benchmarks that report 12 cores ever since a recent HarmonyOS update but this is unlikely) containing four ARM Cortex A510’s running at 1530 MHz paired with three Taishan cores running at 2150 MHz and one Taishan v120 core running at 2620 MHz.

The Taishan cores are
Please, Log in or Register to view URLs content!
processors and the version numbers provide some insight into the revision number printed on the chip. The A510 itself is an ARM Cortex processor from 2020, since superseded by the A520.
 
Last edited:

tonyget

Senior Member
Registered Member
Snapdragon-8-Gen-3-Benchmarks.jpg

Kirin 9010 (HBN-AL00) and Qualcomm’s Snapdragon 8 Gen 3 / Credit: Geekbench

Made in China…Mostly​

It’s notable to see that the Huawei Pura 70 is close to becoming China’s poster child for self sufficiency. The majority of the chips we found are designed and manufactured in China with a couple of notable exceptions.

SK Hynix was notably present, despite claiming that they had
Please, Log in or Register to view URLs content!
since sanctions kicked in back in 2020. It’s very possible that these chips are the last of a dwindling stock of SK Hynix IC’s.

The other big name presence on the PCB was Bosch, with their 6-axis gyroscope and accelerometer MEMS sensor. As a global leader in the supply and design of MEMS sensors, Bosch’s presence would not raise any eyebrows were it not for the fact that Chinese manufacturers probably have the ability to produce these sensors locally.
You can find our
Please, Log in or Register to view URLs content!
on our website.

NAND Flash Memory​

But perhaps the most interesting IC inside the Pura 70 Pro is HiSilicon’s 1TB NAND flash memory. It’s worth noting that the European markets can only get the 256GB and 512GB NAND modules.

NAND-package-top-1.jpg

What’s so interesting about this package? Our friends at
Please, Log in or Register to view URLs content!
provided us with an X-ray while TechSearch provided analysis to help unpack this little beauty.
NAND-xray-side.jpg


Eight NAND die on top of a flip chip-attached memory controller die / Credit: Creative Electron & TechSearch
From the analysis provided, we think it’s possible that HiSilicon may have produced the memory controller and performed the testing and packaging of the NAND chip but a lack of markings on the memory controller itself means we can’t be sure.

NAND-controller.jpg

Memory controller found in the 1TB NAND package

It’s very unlikely that HiSilicon would be able to produce the NAND dies themselves. Again, we’ve been unable to locate any markings on the dies to positively identify the manufacturer but it’s highly likely that they were designed and made locally by another Chinese manufacturer.

NAND-IC-die.jpg

A close up of one of the NAND dies

It’s All About Yield​


There’s no question that China has the expertise to become a self-sufficient semiconductor powerhouse. At least in theory. The immediate bottleneck to advancement is their lack of access to advanced photolithography systems.

It’s fully expected that China will be able to use multi-patterning techniques with their existing DUV (Deep Ultraviolet) photolithography machines to reach an N5 equivalent node. This will happen and there will likely be a wave of alarmism that will follow. However, the shortages and backlogs on orders for Huawei’s
Please, Log in or Register to view URLs content!
and even the months-old
Please, Log in or Register to view URLs content!
imply that, as analysts had predicted, the N+2 process does not produce sufficient yields to meet demand. This restricts SMIC’s ability to produce the 7nm Kirin 9000S and 9010 SoC’s in large enough quantities to meet demand.

Adding to this bottleneck is Huawei’s inability to source LPDDR5 chips from their usual suppliers in the international market due to US sanctions. It seems that so far the Chinese smartphone manufacturer has been able to get around this by relying on a stockpile of chips purchased prior to the 2020 sanctions going into effect. However, those stocks must be dwindling at a rapid rate.

This is where CXMT comes in. As China’s premiere DRAM fab, they’ve recently produced the first fully Chinese-manufactured LPDDR5 module which we expect to start seeing in the wild soon. However, DRAM also relies heavily on the latest node processes for performance and power efficiency. Again, as CXMT attempts to produce on 7nm and 5nm nodes, they will have to grapple with the problem of production yields dropping significantly.

Looking Past 5nm​


For now, the US has successfully restrained China’s ability to mass produce on the 7nm node at high yields. By extension, it has also prevented China from mass producing on an N5 equivalent node at high yields. And yet this has not prevented China from progressing in other areas, notably as a market leader and innovator in 3D NAND technology.

The biggest challenge currently facing the Chinese semiconductor sector is their inability to access the latest node processes. The EUV photolithography machines that enable economical and scalable production of N3 nodes and beyond are exclusively manufactured and sold by a single company in the Netherlands: ASML. While US sanctions remain in effect, Chinese fabricators will be forced to deal with lower yields and higher costs for older tech while simultaneously pouring money into R&D to develop their own EUV and DUV photolithography systems.

Despite these challenges, China is expected to make the leap from a 7nm node process to 5nm imminently, though production yields are likely to remain low. Meanwhile, CXMT (ChangXin Memory Technologies) is predicted to replace SK Hynix as pre-sanctions stocks of LPDDR5 dwindle.

We expect to see a 5nm node from these fabs. The question is, where will they go from there? Rest assured, we’ll be keeping our eyes—and our screwdrivers—on Huawei.
 

MortyandRick

Junior Member
Registered Member
Snapdragon-8-Gen-3-Benchmarks.jpg

Kirin 9010 (HBN-AL00) and Qualcomm’s Snapdragon 8 Gen 3 / Credit: Geekbench

Made in China…Mostly​

It’s notable to see that the Huawei Pura 70 is close to becoming China’s poster child for self sufficiency. The majority of the chips we found are designed and manufactured in China with a couple of notable exceptions.

SK Hynix was notably present, despite claiming that they had
Please, Log in or Register to view URLs content!
since sanctions kicked in back in 2020. It’s very possible that these chips are the last of a dwindling stock of SK Hynix IC’s.

The other big name presence on the PCB was Bosch, with their 6-axis gyroscope and accelerometer MEMS sensor. As a global leader in the supply and design of MEMS sensors, Bosch’s presence would not raise any eyebrows were it not for the fact that Chinese manufacturers probably have the ability to produce these sensors locally.
You can find our
Please, Log in or Register to view URLs content!
on our website.

NAND Flash Memory​

But perhaps the most interesting IC inside the Pura 70 Pro is HiSilicon’s 1TB NAND flash memory. It’s worth noting that the European markets can only get the 256GB and 512GB NAND modules.

NAND-package-top-1.jpg

What’s so interesting about this package? Our friends at
Please, Log in or Register to view URLs content!
provided us with an X-ray while TechSearch provided analysis to help unpack this little beauty.
NAND-xray-side.jpg


Eight NAND die on top of a flip chip-attached memory controller die / Credit: Creative Electron & TechSearch
From the analysis provided, we think it’s possible that HiSilicon may have produced the memory controller and performed the testing and packaging of the NAND chip but a lack of markings on the memory controller itself means we can’t be sure.

NAND-controller.jpg

Memory controller found in the 1TB NAND package

It’s very unlikely that HiSilicon would be able to produce the NAND dies themselves. Again, we’ve been unable to locate any markings on the dies to positively identify the manufacturer but it’s highly likely that they were designed and made locally by another Chinese manufacturer.

NAND-IC-die.jpg

A close up of one of the NAND dies

It’s All About Yield​


There’s no question that China has the expertise to become a self-sufficient semiconductor powerhouse. At least in theory. The immediate bottleneck to advancement is their lack of access to advanced photolithography systems.

It’s fully expected that China will be able to use multi-patterning techniques with their existing DUV (Deep Ultraviolet) photolithography machines to reach an N5 equivalent node. This will happen and there will likely be a wave of alarmism that will follow. However, the shortages and backlogs on orders for Huawei’s
Please, Log in or Register to view URLs content!
and even the months-old
Please, Log in or Register to view URLs content!
imply that, as analysts had predicted, the N+2 process does not produce sufficient yields to meet demand. This restricts SMIC’s ability to produce the 7nm Kirin 9000S and 9010 SoC’s in large enough quantities to meet demand.

Adding to this bottleneck is Huawei’s inability to source LPDDR5 chips from their usual suppliers in the international market due to US sanctions. It seems that so far the Chinese smartphone manufacturer has been able to get around this by relying on a stockpile of chips purchased prior to the 2020 sanctions going into effect. However, those stocks must be dwindling at a rapid rate.

This is where CXMT comes in. As China’s premiere DRAM fab, they’ve recently produced the first fully Chinese-manufactured LPDDR5 module which we expect to start seeing in the wild soon. However, DRAM also relies heavily on the latest node processes for performance and power efficiency. Again, as CXMT attempts to produce on 7nm and 5nm nodes, they will have to grapple with the problem of production yields dropping significantly.

Looking Past 5nm​


For now, the US has successfully restrained China’s ability to mass produce on the 7nm node at high yields. By extension, it has also prevented China from mass producing on an N5 equivalent node at high yields. And yet this has not prevented China from progressing in other areas, notably as a market leader and innovator in 3D NAND technology.

The biggest challenge currently facing the Chinese semiconductor sector is their inability to access the latest node processes. The EUV photolithography machines that enable economical and scalable production of N3 nodes and beyond are exclusively manufactured and sold by a single company in the Netherlands: ASML. While US sanctions remain in effect, Chinese fabricators will be forced to deal with lower yields and higher costs for older tech while simultaneously pouring money into R&D to develop their own EUV and DUV photolithography systems.

Despite these challenges, China is expected to make the leap from a 7nm node process to 5nm imminently, though production yields are likely to remain low. Meanwhile, CXMT (ChangXin Memory Technologies) is predicted to replace SK Hynix as pre-sanctions stocks of LPDDR5 dwindle.

We expect to see a 5nm node from these fabs. The question is, where will they go from there? Rest assured, we’ll be keeping our eyes—and our screwdrivers—on Huawei.
Lol of course you would post an article like this.

Basically this article in a nutshell is a mental masturbation self congratulatory cope article.

"Hey guys look we succeeded in preventing the Chinese from getting VOLUME production of 5nm chips! Woohoo . This is undoubtedly a much better goal than our previous goal of keeping them above 28nm chip production, so really we won!" Lol

However, the shortages and backlogs on orders for Huawei’s
Please, Log in or Register to view URLs content!
and even the months-old
Please, Log in or Register to view URLs content!
imply that, as analysts had predicted, the N+2 process does not produce sufficient yields to meet demand.
I can get a mate 60 tomorrow either through friends or family in China or AliExpress and have it shipped over in a week. What they talk about months ?!?
 

measuredingabens

Junior Member
Registered Member
Please, Log in or Register to view URLs content!

Lithium tantalate photonic integrated circuits for volume manufacturing

Abstract

Electro-optical photonic integrated circuits (PICs) based on lithium niobate (LiNbO3) have demonstrated the vast capabilities of materials with a high Pockels coefficient1,2. They enable linear and high-speed modulators operating at complementary metal–oxide–semiconductor voltage levels3 to be used in applications including data-centre communications4, high-performance computing and photonic accelerators for AI5. However, industrial use of this technology is hindered by the high cost per wafer and the limited wafer size. The high cost results from the lack of existing high-volume applications in other domains of the sort that accelerated the adoption of silicon-on-insulator (SOI) photonics, which was driven by vast investment in microelectronics. Here we report low-loss PICs made of lithium tantalate (LiTaO3), a material that has already been adopted commercially for 5G radiofrequency filters6 and therefore enables scalable manufacturing at low cost, and it has equal, and in some cases superior, properties to LiNbO3. We show that LiTaO3 can be etched to create low-loss (5.6 dB m−1) PICs using a deep ultraviolet (DUV) stepper-based manufacturing process7. We demonstrate a LiTaO3 Mach–Zehnder modulator (MZM) with a half-wave voltage–length product of 1.9 V cm and an electro-optic bandwidth of up to 40 GHz. In comparison with LiNbO3, LiTaO3 exhibits a much lower birefringence, enabling high-density circuits and broadband operation over all telecommunication bands. Moreover, the platform supports the generation of soliton microcombs. Our work paves the way for the scalable manufacture of low-cost and large-volume next-generation electro-optical PICs.
 

Phead128

Captain
Staff member
Moderator - World Affairs
You know the West lost when the goal post moved from 0% yield at <14nm to them saying you have "only" 70% yield at <7nm. As if the goal was to prevent 100% yield to begin with? No, it was 0% yield as the original goal. That's some mental gymnastics right there. It's not worthy to post such trash articles.
 
Top