Semiwiki on LinkedIn Semiwiki on Facebook Semiwiki on Twitter Register / Log in 
Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/tsmc-attacks-silicon-photonics.18758/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021770
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

TSMC attacks silicon photonics!

Daniel Nenni

Daniel Nenni

Admin
Staff member
1694526154777.png


It is rumored that it will join hands with Broadcom, Huida, etc. to jointly develop and welcome large orders next year.
feed

Aiming at the new AI science and technology, an advanced research and development team of more than 200 people was formed to join hands with Broadcom and Huida to seize the ultra-high-speed computing chip business opportunity.

TSMC Telex has joined hands with major customers such as Broadcom and Huida to jointly invest in the application of ultra-high-speed computing chips based on silicon photonics processes, and will begin to receive large orders as early as the second half of next year. (Associated Press)

TSMC Telex has joined hands with major customers such as Broadcom and Huida to jointly invest in the application of ultra-high-speed computing chips based on silicon photonics processes, and will begin to receive large orders as early as the second half of next year. (Associated Press)

AI has created a huge demand for data transmission. Silicon photonics and co-packaged optical components (CPO) have become new technologies in the industry. TSMC (2330) is actively rushing into it. It is reported that it will join hands with major customers such as Broadcom and Huida to jointly develop it, starting as early as the second half of next year . In anticipation of the big order, TSMC has invested more than 200 people to form an advance research and development team, targeting the business opportunities of ultra-high-speed computing chips based on silicon photonics processes that will come next year.

Regarding related rumors, TSMC stated that it does not respond to customer and product status. However, TSMC is highly optimistic about silicon photonics technology. TSMC Vice President Yu Zhenhua recently publicly stated, "If we can provide a good silicon photonics integrated system, we can solve the two key issues of energy efficiency and AI computing power. This will be a new one." Paradigm shift. We may be at the beginning of a new era."

Silicon photonics is a hot topic in the industry at the recently concluded "SEMICON Taiwan 2023 International Semiconductor Equipment Exhibition". Semiconductor giants such as TSMC and ASE have delivered special lectures on related topics. The main reason is that AI applications are blooming everywhere, and how to make huge data transmission speeds faster Problems such as high speed and no delay in reaching signals have come to the fore. The traditional method of using electricity as a signal transmission method is no longer sufficient. Silicon photons convert electricity into light with faster transmission speeds, and have become highly anticipated by the industry to improve the transmission of huge amounts of data. New generation technology for speed.

International semiconductor industry players such as TSMC, Intel, Huida, and Broadcom have successively launched their silicon photonics and co-packaged optical component technology layouts. Explosive growth in the overall market can be seen as soon as 2024.

It is reported in the industry that TSMC is working with major customers such as Broadcom and Huida to develop new products such as silicon photonics and co-packaged optical components. The process technology is extending from 45 nanometers to 7 nanometers. There will be good news as soon as 2024. Entering the stage of mass production in 2025, it is expected to bring new business opportunities to TSMC.

According to industry insiders, TSMC has organized about 200 advanced research and development teams, and is expected to introduce silicon photonics into computing processes such as CPUs and GPUs in the future. Since the original internal electronic transmission lines have been changed to faster optical transmission speeds, the computing power It will be dozens of times that of existing computing processors. It is still in the research and development and academic stages. The industry is highly optimistic that related technologies will become a new driving force for the explosive growth of TSMC's operations in the next few years.

Industry analysis shows that high-speed data transmission still uses pluggable optical components. As transmission speeds rapidly advance and enter the 800G generation, and enter higher transmission rates such as 1.6T to 3.2T in the future, power loss and heat dissipation management problems will be the biggest. The solution introduced by the semiconductor industry is to integrate silicon photonic optical components and switch application-specific chips (ASICs) into a single module through CPO packaging technology. This solution has begun to be certified and adopted by major manufacturers such as Microsoft and Meta. In the new generation network architecture.

Even though CPO technology has just entered the market, the production cost is still high. As the advanced process advances to 3 nanometers, AI computing will drive the demand for high-speed transmission and further drive the restructuring of high-speed network architecture. It is expected that CPO technology will not be ignored and The necessary technologies will enter the market in large quantities after 2025.

What are silicon photons?
Silicon Photonics technology was launched by Intel in 2010. It combines silicon and laser technology, and converts electricity into light technology to convert data originally transmitted from copper wires into optical fibers with faster and more stable transmission distances. The difficulty lies in how to convert electronic data into light. Due to the relatively high cost and the large size of optical communication transceiver modules, they are mostly used in markets such as data centers and servers.

Currently, in addition to developing silicon photonics technology, TSMC is also developing co-packaged optical modules (CPO), which integrate photonic integrated circuits (PICs) in a packaging mode through silicon process wafers, which can convert electricity into optical signals more quickly. This makes the transfer speed faster. (Reporter Su Jiawei)

money.udn.com

台積攻矽光子！傳攜手博通、輝達等共同開發 明年迎大單 | 產業熱點 | 產業 | 經濟日報

AI掀起巨量資料傳輸需求，矽光子及共同封裝光學元件（CPO）成為業界新顯學，台積電積極搶進，傳出攜手博通、輝達等大客戶共...
money.udn.com money.udn.com
 
Mooredaddy said:
How is it unlikely if they are already ahead and have superior financials in every way. Not certain by any means but i think TSMC keeping a chokehold is decently likely
Click to expand...

I don't think it is a big enough of a market to get TSMC's full attention unless customers push then TSMC will in fact own it. They key is big customer participation, just like packaging. If Apple's name is added to the list it will happen, absolutely.
 
The only challenger here is Broadcom. But Intel has spent many years in silicon photonic and manufacture at scale for many years as well.

and for Nvidia,

Nvidia’s research shows co-packaged photonics operating at only 25Gbps per laser, Intel is already capable of 100Gbps and lower power per bit transferred. Additionally, Intel is demonstrating on chip optical modules are connected through an EMIB style interposer while Nvidia is still researching these on chip signals being sent through an organic package. While Nvidia recognizes the problem, their research is already behind what Intel can manufacture. Furthermore, Nvidia has no way of getting what they demonstrated mass produced. (In 2021)

www.semianalysis.com

Intel’s Trojan Horse into the Foundry Business | Co-packaged Silicon Photonics is Intel’s Path Forward for IDM 2.0

Pat Gelsinger’s first move as the new CEO was to do a complete 180 on Intel’s historic business model and enter the era of IDM 2.0. He wants to open up Intel’s fabs to external designs and license …
www.semianalysis.com www.semianalysis.com
Click to expand...

And look, if you heard Pat Gelsinger talking about silicon photonic in Innovation 2022, there's a clip from him 20 years ago predicting SiPh will be the future.

I don't believe he will let Intel's advantage slip by any means.

And how? by acquiring GlobalFoundries or partnering with GlobalFoundries, just like how they did with Tower Semi.
 
Last edited:
Daniel Nenni said:
I don't think it is a big enough of a market to get TSMC's full attention unless customers push then TSMC will in fact own it. They key is big customer participation, just like packaging. If Apple's name is added to the list it will happen, absolutely.
Click to expand...
But Silicon Photonics only matter in data center, right? Apple has been all in SoC, will silicon photonics see any usage there?
 
Mooredaddy said:
How is it unlikely if they are already ahead and have superior financials in every way. Not certain by any means but i think TSMC keeping a chokehold is decently likely
Click to expand...
Intel used to have solid financials for the past decade and more, they spent much more in R&D recently as well. And their past income and cash flow dwarf either Broadcom or Nvidia.
 
delong.height said:
Intel used to have solid financials for the past decade and more, they spent much more in R&D recently as well. And their past income and cash flow dwarf either Broadcom or Nvidia.
Click to expand...

It's true that Intel spent a lot of money in R&D but it might not be true Intel is outspending its competitors in each technology or product category.

I suspect that Intel has difficulty to match the combined R&D strength and spending of the fabless ecosystem. It's one of the reasons why the fabless and foundry business are so prosperous today.
 
delong.height said:
But Silicon Photonics only matter in data center, right? Apple has been all in SoC, will silicon photonics see any usage there?
Click to expand...
Technically Apple CPUs are System in Package (SIPs), where an SoC is one of the chips in the package.

In the data center optical interconnects are just plain better, for a lot of reasons, and optical transceivers have been expensive and power-hungry, so it is true that data center interconnects have the most visible current value proposition for Si photonics. But really high speed chip-to-chip data links on circuit boards are also attractive in optical because you may be able to use cheaper board materials with fewer board layers, and get higher throughput and lower latency. (Not to mention reducing EMI issues, I'm told.) If the chip-to-chip dream I've been hearing about for over 15 years comes to fruition there could be a lot of applications outside of data center equipment, including client, vehicles, and mobile.
 
hist78 said:
It's true that Intel spent a lot of money in R&D but it might not be true Intel is outspending its competitors in each technology or product category.

I suspect that Intel has difficulty to match the combined R&D strength and spending of the fabless ecosystem. It's one of the reasons why the fabless and foundry business are so prosperous today.
Click to expand...
yes, if comparing Intel R&D to TSMC, it's not as effective as it can be. But certainly not every penny spend are all going wasted.

On the other hand, if looking at all those ventures (drone, most ridiculous one) under BK and BS, it's expensive, and deviated from the main business they are doing, which is making semiconductors.
 
delong.height said:
yes, if comparing Intel R&D to TSMC, it's not as effective as it can be. But certainly not every penny spend are all going wasted.

On the other hand, if looking at all those ventures (drone, most ridiculous one) under BK and BS, it's expensive, and deviated from the main business they are doing, which is making semiconductors.
Click to expand...

In the R&D spending for product design, Intel is competing against Qualcomm, MediaTek, Samsung, AMD, Nvidia, Apple, Broadcom, Google, Amazon, Tesla, Ampere Computing, Marvell, Microsoft, etc.

In the semiconductor foundry and manufacturing business, Intel's R&D budget is running against TSMC, Samsung, GlobalFoundries, UMC, SMIC, etc.

Among those Intel's competitors they are sharing many benefits from their R&D efficiency due to the fabless and foundry ecosystem collaborations.

Intel is in a very tough position to do R&D all under its large but fragmented R&D budget.
 
hist78 said:
In the R&D spending for product design, Intel is competing against Qualcomm, MediaTek, Samsung, AMD, Nvidia, Apple, Broadcom, Google, Amazon, Tesla, Ampere Computing, Marvell, Microsoft, etc.

In the semiconductor foundry and manufacturing business, Intel's R&D budget is running against TSMC, Samsung, GlobalFoundries, UMC, SMIC, etc.

Among those Intel's competitors they are sharing many benefits from their R&D efficiency due to the fabless and foundry ecosystem collaborations.

Intel is in a very tough position to do R&D all under its large but fragmented R&D budget.
Click to expand...

You know, maybe you are right that fabless and foundry ecosystem collaborations offer more than what Intel, an IDM, can offer.

But there hasn't been enough proof that an advanced IDM, like Intel, couldn't compete at same level as fabless and foundry if Intel were to have a capable leader that's in charge to begin with. It's hard to say that Intel business structure naturally cannot compete with a pure play fabless company and pure play foundry because there's no history of showing that is the case. You may say that to AMD, or any other that don't have a market position as Intel. But Intel was/is just different, given its scale and market position. It's very difficult to judge given Intel has had a bad culture that's in place since BK. Management placed much emphasis on buyback and diversity program, and different ventures that's has little to do with chip making or chip design.

Now with IDM 2.0, I still have a very high expectation that IDM will outperform fabless and foundry. But of course, the vision has changed. Today, being an IDM isn't just about manufacturing chips and selling them to end customers. It's about bringing a complete solution, or a more modular approach, from top to bottom, from wafer, packaging, design, chiplet, software, IP, assembly, and test, all kinds of aspects under one roof, IFS or system foundry.

And also, if you can remember, the cost initiative of saving $3B in 2023, $8B to $10B exiting 2025. A lot of those are R&D for product design that could potentially be saved, but ultimately wasted because Intel had a bad culture that's in place. They weren't lean, but it didn't say anything bad about IDM. If they were lean and management were smart, but still perform worse than fabless + foundry, then yes, it can be said that IDM is not a good place to begin with. But now, it just doesn't seem that's the case. If Intel is going to deliver the first product, lunar lake, at the end of 2024 on 18A on time. Then I shall say, IDM perform better than Foundry because they are more incentivized to deliver better process node than foundry. Foundry is driven by customer demand, etc. Intel just performed worse under bad execution.

Not only that, TSMC, the leader in foundry, or backend so to speak, has only 5.3B USD in R&D, while Intel has spent over 17B. Well, some of it is 5 nodes in 4 years, so that isn't normal. But even before Pat G come to the office, under Bob Swan, Intel spent $13B. AMD now spent 5B in 2022, which skyrockets from previous 1~2B range. Both TSMC and AMD had outperformed Intel in terms of efficiency, but that was just R&D budget that weren't wisely spent. But it does not say that Intel is in "a very tough place to do R&D under its large and but fragmented".
 
You must log in or register to reply here.
Back
Top