Xinyuan semiconductor Zhang Xiang: the industry's first 28nm process ReRAM chip successfully

From the 80-column punch card of a "high-density storage device" invented by IBM at the end of 1920 to the advent of flash memory in 1980, the memory has gone through many forms of iteration in just 60 years.

With the strong demand for memory in smart phones, cloud services and other markets in recent years, the memory market broke out in 2017, becoming the largest segment of IC. At present, it has formed a market of more than 160 billion US dollars, which is mainly composed of DRAM, NAND Flash and NOR Flash.

However, in the process of development, memory is gradually facing some insurmountable technology gaps, such as the huge gap between traditional storage media and memory media in speed, power consumption and density; in addition, the common problem is that they all face the physical limit of continuous miniaturization of the manufacturing process. The fourth generation memory (also known as "new memory"), which can break through the bottleneck of performance and technology, has gradually attracted the attention of the market.

In this context, Xinyuan Semiconductor (Shanghai) Co., Ltd., a new semiconductor storage technology company specializing in the field of ReRAM, was established in Shanghai and developed rapidly. This is a new IDM company that integrates core technology, process, chip design, IP licensing and production services. The company was co-founded by Dr. Zhang Ke and Mr. Zhang Xiang. In the second year of the establishment of the company, it obtained strategic financing composed of Shanghai Joint Investment, Lianxin Capital, Liansheng Venture Capital and Kunqiao Capital. The company's historical shareholders also include KPCB, Northern Lights Venture Capital, Saifu Asia, Broadband Capital, Yuanhe Valley Wind and other well-known funds.

Zhang Xiang, co-founder and CEO of Xinyuan Semiconductor, told Science and Technology Innovation Board Daily that a chip based on ReRAM storage technology developed by the company last year is now in the return stage and will be shipped in the second half of this year. The chip mainly makes use of the security and storage features of the company's new storage ReRAM, which is mainly used for anti-counterfeiting certification, as well as various accessories for mobile devices, home appliances, e-cigarettes and other AIoT markets. At present, we have reached the intention of cooperation with the leading manufacturers in the industry.

The fourth generation memory has gone through 10 years of development. Zhang Xiang believes that the next 5 years will enter an explosive period, and the market pattern of memory will change.

A new type of memory with distinctive features

According to whether the information stored after power off is retained, the memory chip is divided into volatile memory chip: DRAM/SRAM and non-volatile memory chip: NAND Flash/ NOR Flash. Among them, DRAM and NAND Flash account for more than 95% of the storage market.

DRAM technology development path is mainly to enhance the manufacturing process to improve the storage density, although the microprocessor technology has reached the 5nm node, but DRAM still stays between 20nm and 10nm nodes, no matter Meguiar, Hynix or Samsung Electronics, the highest process of its DRAM products is 1znm (about 12-14nm), while in theory, 10nm is the limit of DRAM.

In contrast, NAND Flash is widely used in SSD memory cells, accounting for about 42% of the memory chip, second only to DRAM,. Its characteristic is that it has a higher density, but the read and write speed is not as fast as DRAM. For NOR Flash, which is driven by 5G, electric vehicle, TWS and other applications in recent years, its write performance is not only slow, but also can not achieve higher density.

Due to the limitations of DRAM and NAND Flash as well as the physical characteristics of NOR Flash, it is difficult to break through the bottleneck of performance and density simply by improving the existing memory, and the new memory provides a solution.

"generally speaking, the new memory has two major characteristics. Take the application scene data center as an example. First, the application of the new memory in the data center can read and write as fast as DRAM while providing higher storage density. Second, all the new memory will save the data after power off (non-volatile). After the application of the new memory, the performance of a data center will be greatly improved, while the energy consumption will be reduced by about 30%." Zhang Xiang described it to Science and Technology Innovation Board Daily.

Based on the performance advantages of the new memory, Zhang Xiang believes that from the point of view of its own performance and power consumption, as long as it is a data center in the future, it will certainly choose this (new memory) development path as a supplement to the current storage configuration. "at present, many companies have come to us to express their demand for new memory."

The future industry pattern may be rewritten.

In addition to the application in the data center, the new memory also has many applications in the cloud, IOT side, vehicle regulation, AI and other fields.

In the field of AI, in the traditional von Neumann architecture, the data is extracted from the memory outside the processing unit and written back to memory after processing. However, due to the speed difference between computing components and memory components in existing AI chips, when the computing power reaches a certain extent, the speed of accessing memory can not keep up with the speed of data consumed by computing components, and the addition of computing components can not be fully utilized, that is, the so-called von Neumann "bottleneck", that is, the problem of "memory wall".

On the other hand, in-memory computing can solve the above problems, which embeds the computing into the memory, so that the memory is not only a memory, but also completes the computing task. The integration of storage and calculation has greatly reduced the cost of data access in the process of computing. "in fact, the most important thing to be solved in in-memory computing is performance and power consumption, that is, the energy consumption ratio," says Zhang Xiang.

For in-memory computing, the characteristics of memory often determine the efficiency of in-memory computing, so when the memory with new features appears, it will often drive the development of in-memory computing. Through the integration of storage and computing, the intermediate movement of data is reduced, so as to improve the energy efficiency of the whole AI chip and break the memory wall problem.

The reporter learned that Xinyuan Semiconductor mainly focuses on the ReRAM (resistive memory) track in the new storage industry. at present, Xinyuan will provide three types of storage products: low-density storage products, medium-density storage products, and high-density storage products. The company's business is mainly focused on four areas: data center, storage particles, AI in-memory computing and IoT chip security.

Zhang Xiang revealed that Xinyuan Semiconductor has completed the successful streaming of the industry's first 28nm process ReRAM chip, which will be mass produced in the second half of this year. The chip will apply IoT chip security areas, including e-cigarettes, batteries, fast charging and other industries that need identity authentication chips; for in-memory computing, it will adopt IP licensing; and in the memory particle market, it will face direct competition with some traditional memories.

Zhang Xiang said that the company's low-and medium-density storage products will be produced in cooperation with contract factories, or self-produced, in which density storage products have advantages over NOR Flash, in performance, density, and cost, while high-density storage products have advantages over SLC NAND products in performance, cost, and density. This also means that with the gradual maturity of new memory products, the existing pattern of the memory industry will be rewritten in the future.

It has the first RERAM production line in the world.

It is worth noting that in addition to the types of ReRAM, the new type of memory also includes phase change memory (PCM), magnetic change memory (MRAM) and ferroelectric memory (FRAM),. What are the advantages of RERAM compared with the aforementioned new types of memory?

According to Zhang Xiang, the reading and writing speed of ReRAM is better than that of traditional memory, but it is slower than that of MRAM and FRAM,. The number of reads and writes of ReRAM is about 1 million, which is an order of magnitude more than that of traditional memory, but less than that of MRAM. "the advantage of ReRAM is very obvious. First of all, in terms of density, the density of ReRAM can be very high."

In fact, it is understood that from the cost point of view, MRAM due to the complexity of materials, density bottlenecks and other characteristics, its cost will be higher. The characteristic of PCRAM is that the density can reach the same level as RERAM, but the reading speed of PCRAM is lower than that of ReRAM,. The biggest difference between PCRAM and RERAM is that the yield of PCRAM needs to be improved because of the complex process.

According to industry insiders, the good rate of ReRAM developed by Xinyuan Semiconductor and foundry has reached 70%, and the good rate will be close to 90% in the future.

In addition, the performance of FRAM is similar to that of MRAM, but its read and write speed is better than MRAM, and can maintain low power consumption. The disadvantage of FRAM is that its cost is higher than MRAM, "so it can be used in some very special markets, such as hearing aids."

Zhang Xiang believes that, on the whole, compared with other new types of memory, density and corresponding cost will be the biggest advantages of ReRAM.

"Science and Technology Innovation Board Daily" reporter learned that in the ReRAM track, Xin Yuan Semiconductor has 10 years of technical accumulation and experience. The company has many years of process development cooperation with the global leader FAB, and is the first company to achieve commercial mass production of 28nm.

In addition, Xinyuan Semiconductor adopts the new IDM model, and the company is the first new storage startup in the world to have a ReRAM back-channel production line.

In fact, in the context of the current global chip shortage, chip manufacturers that adopt IDM mode are highlighting their advantages. For Xinyuan Semiconductor, because of its own production line, in terms of production scale, it not only avoids the complete dependence on FAB, but also fits the existing process production line of FAB, and can match the production quickly; at the same time, the company also greatly shortens the cycle of product development and iteration. In terms of the scale of investment, as only the production of the latter process is carried out, the overall scale of investment can be controlled, and the production capacity can be increased in stages. "at present, the production capacity of the company in cooperation with a contract factory can reach 2000 pieces per month under the lowest equipment, and after the completion of the IDM plant, the company's production capacity will reach 30000 pieces per month." Zhang Xiang said.

Zhang Xiang said that the future application field of the new memory will be very extensive. "in the foreseeable market in the future, memory products will be the company's highest revenue business, and the company's AI business will grow very fast. Xin originally hoped to be the best and strongest new memory company in China in the future."