news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

written by | dale

editor | dong yuqing

due to the huge market size, high degree of standardization, and obvious disturbances from upstream and downstream demand, memory chips naturally have the attributes of a bulk commodity and are known as the "barometer of the semiconductor industry."

looking back at the most recent memory chip downturn, which began in mid-2021, global memory chip manufacturers have cut production. the industry generally believes that since the memory chip market usually has a cycle of 3-4 years, its bottom cycle has now passed and is in the process of rebounding.

but beyond the normal fluctuations, the most surprising part is the emergence of a new round of ai wave in this cycle, which has pushed memory chips to break through the market ceiling.

in this "spring" of memory, we met zhou yanyi, senior vice president of silicon motion technology's enterprise storage and display interface solutions (esdi) business group.

he judged: "a round of big data training is likely to take 4-6 months, and the data at each point in the process needs to be recorded... the temporary storage and processing of these intermediate data is a huge opportunity for change."

zhou yanyi, senior vice president of silicon motion enterprise storage and display interface solutions business group

silicon motion technology co., ltd. (nasdaqgs: simo) was founded in 1995 and is the first ic design company in asia to be listed on the nasdaq. it is the global leader in solid-state storage device control chips, with an annual market share of over 30%. its customers include well-known storage manufacturers such as kioxia, micron, and samsung.

the main components of a solid-state drive can be divided into the main control chip and flash memory particles. the solid-state main control chip is responsible for reading and writing data on the flash memory particles through specific algorithms, and can be called the "brain" of the solid-state drive.

before joining silicon motion technology in december 2023, zhou yanyi worked for broadcom, synaptics, etc. and has decades of experience in the semiconductor industry.

the following is a conversation between him and phoenix v live. for ease of reading, it has been edited and polished without changing the original meaning:

"phoenix v live": how do you view the changes that ai has brought to the storage chip industry?

zhou yanyi:the pipeline of ai big models can be divided into several links, starting with the collection of big data and then transferring it to the training stage. many major changes in chips will occur during training, for example, everyone has paid attention to the explosive growth opportunities of computing graphics cards represented by nvidia, but there is more to it than that.

in certain fields, a single round of big data training may take four or six months. during this process, the data generated at each point needs to be recorded, which means that transient states may appear at a certain training time point. each transient state involves many parameters.

therefore, faced with the massive amount of data generated by ai training, how to store it and where to store it become the key.

"as data gets bigger, more storage is required," which is a very reasonable logic.

therefore, in response to this wave of market demand, we need to develop storage products with high performance, high capacity, high energy efficiency and high security.

phoenix v live: so what is silicon motion's product or solution portfolio to cope with this round of ai wave?

zhou yanyi:we have two bus (business units), and the company has always been in a leading position in the consumer target market such as pcs and mobile phones; i am mainly responsible for the enterprise market.

in the tob market, in addition to the products we have previously deployed, this time we brought the montitan™ enterprise-class pcie 5.0 ssd solution platform. this is a type of product that our competitors do not have, but we have it alone.

the platform covers the solid-state drive master control chip code-named "sm8366", as well as enterprise-level firmware and other technologies. it is an integrated hardware and software business platform that allows customers to build different enterprise-level solid-state storage products based on different business needs.

we launched this product by chance at a time when market demand was strong, which was a well-thought-out decision. for example, our leading position in qlc (a flash memory technology architecture with high storage density and low cost) was because we had a global leading storage system manufacturer customer who was exposed to the demand early and completed the research and development and registration of patented technology.

"phoenix v live": what do partners think of silicon motion?

zhou yanyi:high performance, high capacity, high energy efficiency, and high safety are all trends. we have quickly gained recognition from clients in meeting new demands, so we have received a lot of inquiries and hope to have more market share in the future.

currently, the global nand flash (flash memory chip) manufacturers include samsung, sk hynix, micron, etc., as well as china's yangtze memory. we maintain regular communication with these manufacturers to facilitate the advanced layout of technology and define our products by predicting the development direction of the market in the next 3-5 years.

one of the prerequisites for us to maintain good cooperation with these companies is that silicon motion has only focused on making main control chips for decades since its establishment, and has not competed with flash memory particle manufacturers and module manufacturers. this is our core spirit.

phoenix v live: what is the current size of silicon motion's staff and r&d?

zhou yanyi:the team is mainly concentrated in the united states, taiwan, shenzhen and shanghai, with a total of more than 1,600 people worldwide, nearly 90% of whom are r&d personnel.

phoenix v live: in 2022, the us listed chip company mylink launched a takeover offer for silicon motion, but the offer was unexpectedly terminated before the acquisition was officially completed in 2023. what do you think of this? are you considering seeking some capitalization actions in the future?

zhou yanyi:of course, the acquisition is now a thing of the past, but it was originally an accident.

under the institutional framework of the united states, when someone makes a good takeover offer, the target company must consider whether to accept it or not. but internally, selling the company has never been the intention of the company's management. this was an episode, and it did have some impact on us.

however, the industry should have seen our efforts to maintain normal operations over the past year. we have spent time and tried hard to explain to our customers.

in the future, in the context of geopolitical struggles, similar things are unlikely to happen again. in the foreseeable future, our strategy is to focus on operations seriously.