1.*Advantages of NPU Switches**
The NPO-based NPU switch released by Huashan and Ruijie has the advantage of system loss tolerance, and the loss is reduced from 10 DB to 13 DB.
2.Different perspectives on CPO technology:
There are differences in the needs of different industry players for CPU technology, such as Aretta believes that 100TB switches do not need CPU technology, while Huawei believes that 3Optical modules can still be used with 2TB switches.
3.**Industry Consensus for CPO Technology*:
The industry consensus is that CPO technology is at least 32TB switches, and it may take 4 to 5 years for large-scale commercial use.
4."The Role and Future Development of Silicon Photonics**:
As a third-party platform provider, TSMC has the opportunity to occupy a place in the field of silicon photonics technology, and it is expected that silicon photonics modules may account for 50% of the global optical module market by 2026.
5.*Silicon photonics technology and industrial chain development**-Silicon photonics technology has matured, the industrial chain is becoming more and more perfect, and the cost is more advantageous than traditional solutions. 400g silicon photonics has been tested by Amazon, and 800g silicon photonics modules are also expected to get orders this year.
6."Silicon photonics module market penetration is expected:
The proportion of the silicon photonics module market will not be particularly high this year, but there may be explosive growth next year and the year after, and it is expected that it may occupy half of the optical module market in the future.
7.*Business model and reliability of CPO technology*:
Although the CPO technology is good, there are uncertainties in the business model, and it will take time for large-scale commercial deployment. The defective rate of silicon photonics chips needs to be reduced to 9999% to improve reliability.
8.*800G optical transceiver market penetration rate and technical route:
The market penetration rate of 800G optical modules is expected to be about 10% this year and may reach 50% in the next three years. The technical route includes interface technology and silicon photonics 20, facing challenges including silicon photonics chip selection and customer certification verification.
9.*Silicon photonics chip market situation and TSMC's role:
Intel is the leader in silicon photonics chips, and TSMC has cooperated with Google to gain advantages in the development of 800G optical modules, and is one of the main leading companies.
10.**Testing certification and mass shipment in the field of silicon photonics chips*:
TSMC and InnoLight have partnered to promote testing and certification, and in the face of high costs and packaging threshold challenges, TSMC may continue to promote the use of self-developed solutions.
11.**Market opportunity for silicon photonics in the 800g era*:
Silicon photonics technology has opportunities in the 800G era, even if EML and WAX are not out of stock, silicon photonics has market opportunities in the 800G era.
12.**The growth rate of the 800G optical module market is expected.
The demand for 800G optical modules is expected to be between 10 million and 15 million, and the market growth rate may be about 50% per year in the next few years.
13.**Cost comparison of silicon photonics 800g versus traditional transceivers. *:
Silicon photonics 800G modules can maintain at least 20% of the best advantage under the same industrial chain cost, which may provide higher gross profit margins.
14.**Investment Prospects for Silicon Photonics Technology**:
Silicon photonics solutions are the future technology direction, but the market has not yet formed a clear demand, which is uncertain. Silicon photonics technology is competitive in terms of performance and power consumption, but faces challenges in reliability and testability.
15.**Investment Trend of Silicon Photonics Module Industry**
Investors should pay attention to the trend of market share and value concentration to leading companies, especially those with cost advantages in the field of silicon photonics modules.
This document provides an in-depth analysis of the silicon photonics technology and 800G transceiver market, including technology developments, market expectations, cost-advantage, investment outlook, and industry trends.
O&aq: What are the advantages of the NPU switch released by Huashan and Ruijie based on NPO in terms of tolerating loss on the system?
A: The NPO-based NPU switch released by Huashan and Ruijie has a greater loss on the system, which was originally reduced to 10 DB, but now it only needs to be reduced to 13 DB, which provides a higher tolerance for the system. Q: What are the different voices and perspectives in the industry regarding Co-PackagedOptics (CPO) technology? A: Different industry players have different requirements for CPU technology, for example, Aretta believes that CPU technology is not needed at all in 100TB switches, while Huawei may be more aggressive, thinking that even 32TB switches can also continue to use optical modules.
Q: What is the general consensus in the industry on the application of CPO technology?
A: The industry consensus is to be at least 3CPO technology can only be used on 2T. Facebook and Broadcom are already working in the field of switch solutions with higher bandwidth, while CPO technology could be 4 to 5 years away if it is truly commercially available at scale.
O: What is the company's role and future development in silicon photonics technology?
A: As a third-party platform provider, TSMC has been led by overseas giants such as Intel and Cisco in silicon photonics technology in the past. In recent years, with the maturity of technology and the improvement of the performance of silicon photonics chips, TSMC also has the opportunity to occupy a place in this field. The technology and industry chain of silicon photonic modules are becoming more and more perfect, and it is expected that by 2026, it may account for 50% of the world's optical modules.
Q: What is the current development of silicon photonics technology and industrial chain on the market?
A: Silicon photonics technology has matured, and the industrial chain is becoming more and more perfect. In terms of cost, the cost of using silicon photonics has more than 20% advantages over traditional solutions. 400G silicon photonics has been tested on Amazon, and it is expected that there will be orders in the second half of this year, and 800G silicon photonics modules have also been tested, and it is possible to get orders this year.
Q: What is the expected penetration rate of silicon photonics modules in the market?
A: Although the market share of silicon photonics modules will not be particularly high this year, there may be explosive growth next year and the year after. For example, domestic leading optical module companies such as InnoLight have begun to deploy silicon photonics, which has increased from 800g to 1The shift in the 6t direction can be seen. It is expected that silicon photonic modules will occupy half of the optical module market in the near future.
O: What is the consensus in the industry regarding the business model and reliability of CPO technology?
A: Although CPO technology is a good concept, in terms of business model, there is still a long time before it can be deployed on a large scale due to many uncertainties. The defect rate of optical chips on the market is about 3% to 5%, and for CPO application, the defect rate of silicon photonic chips needs to be reduced to 9999% to improve reliability. At present, the consensus reached in the industry is 1CPO will not be widely deployed before 6T, but from 16t to 3There is still a lot of potential for the development of 2t.
Q: What is the company's market penetration expectation for 800G optical modules? What are the current technical roadmaps and challenges?
A: At present, the market penetration rate of 800G optical modules is still growing, and the penetration rate is expected to be about 10% this year. In the next three years, the 800g silicon light penetration rate may reach 504And 1The demand cycle of 6T optical modules will be a few years later, but it will be around 2027, and it is expected to be 1The proportion of 6t can reach more than 50m. At present, there are two main solutions for 800G technology: one is interface technology, which has a relatively low yield rate and is not mass-produced; The second is silicon photonics 20, that is, silicon photonic chips that integrate some other materials. The advantage of silicon photonics 20 is that the cost and yield may not be much different from that of single-wave 100G, but the supporting chip is not yet perfect, and it is still in the research and development stage. 1.Once 6T silicon photonics is deployed in batches, the cost advantage will be further highlighted. The selection and verification of silicon photonics chips and customer authentication are current challenges, which require rapid technology selection and continuous client verification.
Q: What about the market situation of silicon photonics chips and the role of TSMC?
A: The leader of silicon photonics chips is Intel, whose model is different from that of other companies, and is currently transforming into the role of a leading supplier of silicon photonics chips. Other major players include Acacia and Marvel, but they have encountered challenges in promoting 400g silicon photonics solutions. Due to the cooperation with Google, TSMC has achieved advantages in the development of 800G optical modules in the early stage, and is now the main leader in 800G modules, and the certification of silicon photonics solutions on the client side is progressing relatively quickly. TSMC has shipped hundreds of thousands of silicon photonic modules last year, mainly to first-class customers, and has accumulated a large number of experience in mass manufacturing of silicon photonic modules. TSMC is expected to have a large market share in the 800g silicon photonics market in the future. Some other domestic companies are relatively slow in the certification of silicon photonics modules, and 800G silicon photonics is a stage for TSMC to increase its volume.
Q: What is the progress of the silicon photonics chips in China?
A: Xinxingsheng has passed Amazon's test in terms of 400G, and it is expected to start receiving orders from the official website this year. However, for 800G silicon photonics chips, it seems that they have not been tested yet, and the development progress is relatively slow.
Q: After the acquisition of Alpam Company, what is the company's strategic layout for silicon photonics chips? What are the differences between other companies in the selection of silicon photonics chips?
A: TSMC's acquisition of Alpam has really strengthened its silicon photonics chip business. InnoLight and Xinyisheng rely on self-developed chips, and for different companies, their strategies are diversified, such as Acacia also has its own silicon photonic chips. Enterprises generally do not bind a single solution to reduce risk. At present, even if TSMC's self-developed solution has passed the test, it is still possible to use Acacia's solution on some product models. However, it should be noted that the overall cost of Acacia's chips is relatively high, coupled with the threshold of chip packaging technology, there may be difficulties in promoting it in the future.
Q: What is the progress of the company's testing and certification in the field of silicon photonics chips? In the face of the challenges of high cost and packaging threshold, what is TSMC's response strategy?
A: TSMC has a good cooperative relationship with InnoLight, which has promoted a series of tests and certifications for InnoLight. Although DSMC faces challenges such as high costs and packaging technology thresholds, TSMC is likely to continue to promote the use of self-developed solutions because the test of the self-developed solution has passed. As for cooperation with other companies such as Acacia, its solution may be adopted on specific product models, but considering the cost, TSMC may not widely promote Acacia's chips.
Q: What is the company's batch shipment in the field of silicon photonics chips? Compared with Xu Chuang, Xinsheng and other companies, what is the level of Yunhui? A: Yunhui did not achieve batch shipments in the silicon photonics chip leader, they used the INSIDE solution, mainly multi-mode optical modules, ranking among the top in the world. Compared with other companies, Yunhui has not set foot in the single-mode market, and only provides optical module design and packaging services in the field of silicon photonics, and the cost of silicon photonics chips is high and the packaging is complex, which may be a factor affecting the batch size. Although some products were shipped later, they did not reach large-scale batches.
Q: After Yunhui was acquired by Lumen, did the acquisition bring integration advantages in terms of CW light source and silicon photonics chip design? A: Yes, the halogen door has the advantage of batch shipment in CW light source, and its quality is relatively good. At present, silicon photonics chip integration looks like a more feasible commercialization path. Lumen has the ability to empower Yunhui's integration in the field of silicon photonics chips, providing Yunhui with integration advantages.
0: For 16T silicon photonics module, is it necessary to use thin-film lithium phosphate?
A: In the current situation, if you want to achieve a single wavelength of 200G, pure silicon photonics technology using SO1 materials still has limitations. Therefore, the use of thin-film lithium phosphate appears to be a proven solution. The natural advantage of thin-film lithium phosphate bandwidth can stably provide a single wavelength of 200g or even 400g, which provides expansion potential for silicon photonics technology. There are indeed not many manufacturers of thin-film lithium phosphate in the world, and the maturity of the technology needs to be improved, but with the emergence of market demand, these technologies will progress rapidly. Breakthroughs in silicon photonics technology can take time and a lot of engineering research.
Q: As the speed increases, will server-side CPUs become an option for silicon photonics technology due to performance limitations?
A: With the data center switch from 512T to 100T, and there may be a need to support single-wave 200G rates in the future. When 200G is reached, the advantages of CPUs are even more pronounced in terms of signal integrity, where silicon photonics solutions may become a must. But for now, in the next five years, this demand may not emerge in a hurry, because it is currently 51The 2T processor has not yet been deployed on a large scale.
Q: What is the company's current market penetration rate of silicon photonics technology and its opportunities in the future EML shortage? What is the current state of development of silicon photonics technology, and what is the result of Intel's early investment in this technology?
A: Silicon photonics technology has not shown its advantages in the 100G era, and its advantages are not obvious in the 400G era. However, in the 800G era, silicon photonics technology has reached a critical juncture, the technology has become relatively mature, the cost structure has become transparent, and the cost of optical modules can be saved by more than 20%. Now, large companies like Google and Nvidia are willing to test silicon photonics solutions. Even if EML and WAX are not out of stock, silicon photonics technology itself has an opportunity in the 800G era. If silicon photonics reaches a comparable shipment volume, it has huge room for growth in the future. Although silicon photonics' market share may be less than 10% this year, if its shipments can reach the million level, the cost of its chips may be similar to that of EML when it is mass-produced, which provides hope for potential market development in the future. It must be clear that silicon photonics technology does not need to be out of stock in EML to have a chance, it already has its own market opportunity in the 800G era. To achieve a breakthrough in the market, silicon photonics needs to reach hundreds of thousands of shipments this year.
Q: What is the current market situation of 800g products? In terms of technology selection, what is the timing and preparation of silicon photonics for the future market?
A: The market growth window for 800G products is expected to continue for a long time, driven by the demand for A1 technology and the upgrading of data centers, which are generally likely to start a large-scale switch to 800G around 2025. The industry generally believes that 800G will become the mainstream node, and the life cycle of 800G may be at least five years, and the entire cycle may last 7 to 10 years. In this cycle, silicon photonics still has the opportunity to lay out the market. But the condition is that silicon photonics needs to reach hundreds of thousands of shipments this year, otherwise the large-scale application in the 800G market will be affected.
Q: For those new players who have not yet entered the 800g market, how should they capitalize on the opportunities?
A: If new players do not deploy 800G in time and do not enter the ** chain of the data center, it will be very difficult to enter the market. For companies that are already in the chain, customers will usually be offered certification opportunities, but if they are not in the chain at all, their opportunities are unlikely to be more in the data center upgrade and the need for new technology routes.
Q: What is the company's future investment prospect in silicon photonics solutions?
A: Silicon photonics solution is indeed the future technical direction, but the market has not yet formed a clear demand, so there is great uncertainty Although silicon photonics technology is competitive in terms of performance and power consumption, at present, everyone is on the same level. Silicon photonics involves optoelectronic devices, which are technically difficult and face challenges in terms of reliability and testability. Although TSMC has relevant research, it is more of an exploratory work. In addition, in the current pattern of the 800G optical module market, manufacturers that have not yet been shortlisted may be more passive. Therefore, the development of silicon photonics solutions is subject to further changes in the market and demand confirmation. Q: What is the expected market growth rate of 800G optical modules?
A: The current demand for 800G optical modules is expected to be between 10 million and 15 million, which is several times higher than the previous year. Judging from the shipment of A1 chips and the upgrade of data centers, the market growth rate in the next few years, conservatively estimated, may increase by about 50% per year. If the demand for AI explodes, the demand growth rate is likely to accelerate further.
Q: How does the silicon photonics 800g compare with traditional optical modules in terms of cost, downward trend and gross profit?
A: It is expected that the ** of 800G optical modules may be reduced by 20% from last year, and with the maturity of **, it may decrease by 10% to 12% per year, which is an acceptable level for **chain customers. Compared with traditional optical modules, silicon photonics maintains at least 204 advantages under the same industrial chain cost. The cost of silicon photonics may now be similar to that of traditional solutions, but due to its higher power consumption, the main competitor is the PagePulse solution. Silicon photonics solutions are likely to provide higher gross margins, especially with larger shipments. In the short term, the gross profit of silicon photonics may be about 10% higher than that of traditional optical modules, but this also depends on the pricing strategy and chain capacity of the supplier and the customer.
Q: What is the current cost split of 800G silicon photonic modules?
A: The cost of 800G silicon photonics chips can be reduced to less than $40, based on the volume of 100,000 units. CW lasers cost about $17 each, and depending on the design, two to four are required. Traditional transducer and receiver solutions cost about $35 to $40, while drivers and TIAs cost about $60 to $80. DSPs now cost more and can cost around $120. Add in PCBA and mixing costs, and the overall cost of silicon photonics can be around $320. If you can produce silicon photonics chips by yourself, the overall gross profit will be higher.
Q: Is the company's 200G optical transceiver chip technology mature?
A: At present, although single-wavelength 200G optical module chips have been reported to achieve certain performance in academic institutions, and can be close to such a level at the laboratory level under special conditions, there are still difficulties in engineering and mass production. At present, there is only one company in the world that can mass-produce single-wavelength 100G optical transceiver chips, so it is still difficult for 200G technology to achieve engineering and mass application. The industry generally believes that whether it is silicon photonic modules or other technologies, reaching 200G in physical performance is very limiting.
Q: What trends should investors pay attention to in the optical module industry?A: The pace of technology iteration is very fast, and the competition in the entire optical module industry is fierce, but the trend that can be seen is that the market share and value volume will tend to be concentrated in the head enterprises. This should be paid special attention to when investing, especially in the field of silicon photonics, which is considered to be a trend that could disrupt the entire industry landscape. Silicon photonics has a 20% cost advantage, especially for those leading companies that have been tape-out at TSMC since 2017 and 2018. At present, these enterprises have reached a certain maturity and yield rate, which will further show their cost advantages.