Alibaba Dharma Academy successfully developed a storage-computing integrated chip; SiFive announced the new RISC-V processor P650, which has a performance that exceeds the previous generation by 50%...
Alibaba P650
On December 3, Alibaba Dharma Academy announced the successful development of a new architecture chip. This is the world's first DRAM-based 3D bonded stacking storage-calculation integrated AI chip. It integrates a number of innovative technologies and can break through von Neumann. The performance bottleneck of the architecture, in certain AI scenarios, the performance of the chip is increased by more than 10 times, and the energy efficiency ratio is increased by more than 300 times.
Carol Comments: With the development of technology, computing tasks are becoming more and more complex, and more and more data are needed. In the Von Neumann architecture, data needs to be continuously moved in storage, memory, cache, and computing units, resulting in Most of the time, bandwidth, cache, and power consumption are spent on data handling, not calculations, so the memory wall has become an increasingly serious problem. This kind of problem is especially obvious in artificial intelligence computing. And storage-computing integrated architecture chips can be a good way to solve this problem. In recent years, more and more companies have joined the research of storage-computing integrated chips, including Zhicun Technology, Homo Smart, Pingxin Technology, and Zhixin Technology. Some are AI startups, some are storage companies, such as Qianxin Technology, Gencun Technology, and Shine Semiconductor. Only a few companies have achieved mass production of integrated storage and computing chips, such as Zhicun Technology.
The integration of storage and calculation mainly includes near-memory calculation and in-memory calculation. In-memory calculation is more efficient than near-memory calculation. In-memory calculation generally uses stored parameters to complete the calculation, such as DRAM, SRAM, Flash, RRAM, etc. It can be seen that Dharma Institute adopts the AI chip design based on DRAM to realize in-memory computing, while Zhicun Technology uses Flash. At present, the research directions of each company are slightly different. This is a good thing, because in the early stage of new technology development, The more directions you try, the more likely it is for the entire industry to succeed in the future.
GF is hot, and a large number of irrevocable orders supports its high valuation
This week GF released its first financial report after its listing. The report showed that its third-quarter revenue increased by 56% year-on-year to US$1.7 billion, and its fourth-quarter revenue is expected to increase by 71%. Moreover, it is working hard to expand its production capacity and plans to more than double its capital expenditure next year.
Kevin commented: Since its listing at the end of October, GF has been very popular. Before its third-quarter earnings report, its stock price has risen by 47% from the IPO price. Therefore, the company’s valuation has now ranked among the top of the semiconductor sector. If the expected profit is used to calculate the price-earnings ratio, it is already close to 78 times, which is 3 times the price-earnings ratio of TSMC.
How high GF can get a valuation is inseparable from the current chip shortage. Due to the shortage of chips, some key products used in smart phones, data centers and automobiles have received market attention. These chips are basically mature technology, and this is precisely where GF's advantage lies. At present, the capacity utilization rate of its chip manufacturing facilities has reached 100%.
At the same time, the chip shortage also allowed GF to effectively lock in business for the next few years. According to reports, the company now holds more than $20 billion in non-cancellable long-term agreements, which is almost four times its annual revenue in previous years. Moreover, its CFO also stated on the financial report analysis conference call that a large part of these orders are "single-source" procurement transactions, which basically offsets the cancellation of orders by chip buyers because they find supply sources elsewhere. risk.
It is precisely because of a large number of orders to lock in its future production capacity that GF's valuation has become higher.
Mass production of the world's first aluminum-ion solid-state battery
According to reports, Asian battery research and development company Saturnose announced that it will publicly release independent test results of enhanced aluminum-ion batteries, and plans to launch a solid-state rechargeable aluminum-ion battery, which will be released as soon as next year. This may be the world's first mass-produced commercial-grade aluminum-ion solid-state battery, which is expected to replace the risky lithium-ion battery.
Felix comments: From the perspective of technology development, raw material reserves and safety, the possibility of lithium-ion batteries being replaced in the future is very high. One of the alternatives is sodium batteries. This new type of battery with many advantages, such as stable properties, high safety, and long service life, is feasible to replace lithium batteries. Considering that the earth's sodium reserves are 406 times that of lithium reserves, this substitution is greatly increased.
Of course, the new mainstream solution has not yet been determined, so aluminum-ion solid-state batteries are also an alternative. The aluminum-ion solid-state battery is expected to provide 600Wh/kg of energy, which is much higher than the 200Wh/kg of the ternary lithium battery, and the battery life is expected to double. The more important point is safety. According to Saturnose, the production of aluminum-ion solid-state batteries does not use nickel and cobalt, and there are no lithium-ion dendrites and thermal runaway fire problems. Therefore, according to the data published on the official website, aluminum-ion solid-state batteries not only have higher energy density and are safer, and have great alternative advantages, but the actual measurement situation depends on the situation after the subsequent products are launched.
SiFive announces the new RISC-V processor P650, the performance exceeds the previous generation by 50%
American RISC-V company SiFive recently released the latest RISC-V processor P650, which uses a thirteen-stage pipeline 4 launch configuration and supports up to 16 cores. According to SiFive, its performance is more than 50% higher than the previous generation P550 processor, and it has a significant PPA advantage compared to ARM Cortex-A77.
Leland comment: RISC-V ushered in a new processor for high-performance applications. It is not clear how this processor compares with the flat-head Xuantie 910 launched by the 12-level pipeline. However, SiFive claims that this is the fastest licensable RISC-V processor IP core on the market. On the basis of the previous generation P550 processor, SiFive has increased the instruction issue bit width and achieved a 40% performance improvement per clock cycle. . At the same time, the P650 has carried out additional structural optimizations, increased the maximum clock frequency, and finally achieved a performance improvement of more than 50%. SiFive engineers estimated that its test running score was above 11SPECInt2006/GHz.
At the same time, RISC-V International has also recently approved 15 new specifications, providing new opportunities for RISC-V in markets such as AI, machine learning, IoT, connected vehicles and autonomous driving, and data centers. However, given that there is still a certain gap between the P650 and Arm mainstream mobile processor cores, not to mention that the P650 will not be officially released until next summer. If it is to be benchmarked at the same time, the competitor is ARM's Cortex-X2. From this point of view, Smart phones are probably still the most difficult market for RISC-V to squeeze into.
Multiple factors caused a significant decline in domestic robot growth in the third quarter
According to MIR statistics, China's robot shipments in 2021Q3 increased by 28.7% year-on-year, and the year-on-year growth rate has slowed down significantly compared with the 96.3% and 75.8% levels in the first and second quarters. In addition, according to the National Bureau of Statistics, the output of China's industrial robots from January to October increased by 51.9% year-on-year to 298,100 units. Looking at the situation in a single month, the output of China's industrial robots in October was 28,500 units, a year-on-year increase of 10.6%, and the growth rate was 8.9 percentage points lower than that in September, continuing the trend of a sharp slowdown.
Sisyphus comment: Multiple factors have combined to cause a significant slowdown in robot output year-on-year growth. First of all, affected by multiple factors such as tight power supply, chip shortages, and high prices of some raw materials, the supply chain of the robotics industry has been affected to a certain extent, and the slowdown in growth is expected. Second, the domestic manufacturing industry has been below the boundary between prosperity and decline for two consecutive months. In the context of further weakening of the supply and demand ends of the manufacturing industry, the growth of downstream demand in the robotics industry has also been suppressed to a certain extent.
The decline in the growth rate of domestic robot shipments can also show the structural differentiation of industry demand. From the perspective of application areas, the demand for robots in the photovoltaic and lithium battery industries is still very large, and shipments are still increasing. In addition, the market demand in semiconductors, medical supplies, food and beverages and other fields is also relatively strong. These fields are driving robots in the third quarter. The main driving force for market growth. However, the shipment volume of robots in the electronics, manufacturing, and home appliances sectors was significantly lower than the previous average, resulting in a significant slowdown in the overall growth rate of robots.
At the same time, competition in the domestic market has intensified, and many companies in the industry have failed to achieve their expected performance. Under the situation that the growth rate of domestic robots has slowed down significantly, the profitability of the entire industry has declined due to the impact of rising raw material prices.
