If China delivers on its promises, the industry could face an overcapacity of flash memory in the next few years.
Brian Matas, vice president for market research at IC Insights, says companies planning to significantly ramp up 3D NAND flash capacity include Samsung, SK Hynix, Micron, Intel, Toshiba/SanDisk and XMC/Yangtze River Storage Technology and additional new Chinese producers possibly entering the market. Matas wrote: “IC Insights believes that the future risk for overshooting 3D NAND flash market demand is very high and growing.”
__Figure 1:__ *Comparison of major global NAND flash suppliers (Source: DRAM eXchange, March, 2016)*
In recent weeks, EE Times has talked to several semiconductor industry sources, as well as academics in the United States and Chinese analysts, hoping to untangle the web of relationships among China’s memory players, its investment funds, potential technology transfers in the works and non-Chinese tech companies angling for more deals with China.
Let’s start with outfits actually producing memory chips in China today.
Among these are Samsung fabricating 3D NAND flash in Samsung’s fab in Xian, SK Hynix producing DRAM memory at its plan in Wuxi, and Intel—which this year began ramping 3D NAND flash in its converted fab in Dalian.
On the other hand, as IC Insights’ Lineback reminded us, wafer foundry SMIC phased out DRAM manufacturing more than five years ago.
As Lineback puts it, China’s domestic memory chip makers are still tiny. Alarming, however, is the emergence of high-profile memory chip makers in China, financed by the Big Fund, local governments and private equity funds.
According to IC Insights’ Matas, these rising local memory chip vendors include:
Typical of many Chinese enterprises, information about these vendors’ plans remains murky at best.
But Yangtze River Storage Technology, in particular, is worthy of note. Tsinghua Unigroup engineered the deal in response to two foiled efforts to buy U.S. companies.
Tsinghua Unigroup first failed in a ₹1.53 lakh crore ($23 billion) bid for Micron Technology in 2015. Then, earlier this year, its ₹25,331.64 crore ($3.8 billion) plan to become the largest shareholder in data storage group Western Digital came up short.
However, Tsinghua Unigroup’s acquisition of a majority stake in XMC this July 26, and the subsequent formation of a new holding company for XMC called Yangtze River Storage Technology Co. (YRST) is widely viewed as an explicit sign that China has not given up its memory ambitions.
XMC in Wuhan, run by Simon Young, ex-SMIC executive, has been engaged in memory production. It has been producing NOR chips for Spansion (now a part of Cypress Semiconductor). In early 2015, it partnered with Spansion to develop 3D NAND. XMC also makes Nor flash for Beijing-based GigaDevice.
YRST’s stated plan reportedly involves building a 300mm fab in three phases for ₹1.60 lakh crore ($24 billion). First-phase construction has begun, with completion scheduled in 2019.
Reports also say that, using Spansion's technology, YRST’s capacity is expected to be about 300,000 wafers monthly for 32-layer 3D NAND flash chips as early as the end of 2017.
YSRT, however, has its skeptics.
Many Japanese industry sources we contacted remain doubtful of the 3D NAND technology that XMC says it’s got. Even though China, Inc.’s rush to build a memory business is real, just how—and how soon—it can actually make NAND flash remains a complete mystery, they said. In their view, the hitch is the intellectual property behind the 3D NAND technology that XMC claims it’s been developing.
East-West Centre’s Ernst said, “Spansion’s 3D NAND has never seen production, and is considered inferior.”
Moreover, “By the time XMC brings 32 layer 3D-NAND to the market, Samsung and other incumbents may be at 100-plus layers and new memories will be coming onto the market such as 3D Xpoint.”
Dieter cited researcher Sanford C. Bernstein, who famously said that in 2013 China spent more money importing chips than it did importing oil. Bernstein, said Dieter, “estimates XMC will burn ₹6.67 lakh crore ($100 billion), minimum.”
So if it’s unrealistic for YRST to depend on XMC’s 3D NAND, what are YRST’s options?
If it isn’t allowed to get up and buy an accomplished memory chip maker, can China gain access to the core memory technology? Is JV an option? Who will be there to draw a line in the sand?
Entering the rumour mill is word that Tsinghua Unigroup has approached Micron Technology about a licensing agreement to produce 3D NAND flash products.
IC Insights’ Lineback doubts this scenario. “It won't be easy for [China’s] investment groups and Chinese companies to acquire leading-edge DRAM and nonvolatile memory technologies since governments in South Korea, Japan and the U.S. will undoubtedly block deals,” he said. “Even partnerships will be scrutinised.”
He added, “Also on the table is how long NAND flash and even DRAM will be viable memory technologies into the next 10 years. Will China be able to repeat South Korea's success in DRAM after taking the lead in dynamic RAM from Japan? In the last two decades, Taiwan failed to take the lead from the big Korean companies.”
Ernst views licensing agreements as the only viable option left for the Chinese.
“In line with major initiative (through China’s Patent Office, Ministry of Science and Technology, Ministry of Commerce) to improve licensing capabilities of Chinese firms, following the model of Huawei, I would not be surprised if Tsinghua/YRST/XMC have hired the service of patent licensing service companies to explore how to gain access to key 3D NAND patents,” said Ernst.
For example, Micron has a history of selling patents to non-practicing entities (NPE) like Round Rock, said Ernst. Round Rock Research LLC is a N.J.-based technology investigation and patent licensing company whose portfolio holds several thousand issued patents and pending applications throughout the U.S., Europe and Asia.
Ernst pointed out, “Micron is also involved in multiple cross-licensing arrangements. There are specialised IP service companies who could arrange such licensing deals with much discretion. Those companies include Quatela Lynch McCurdy, a Rochester, N.Y.-based IP consultancy, or even Xiaomi’s IP consulting firm Zhigu, he explained.
Ernst said, “Even CFIUS may not be able to detect this, at least in time to prevent it. Nor is it clear whether there are the necessary legal instruments in place to prevent it.”
He concluded: “MXC may sooner or later be able to licence 3D NAND technology. The question then of course becomes: At what price, and at what conditions?”
Lineback is particularly sceptical of Tsinghua and XMC getting 3D flash technology from Micron. Citing 3D Xpoint, a non-volatile memory technology Intel and Micron announced in July, 2015, Lineback said, “I don't think Intel is likely to share its XPoint technology at this stage.”
However, relationships between Chinese partners and big multinationals like Intel are never straightforward. Lineback said Intel announced it would buy a 20% stake in Tsinghua Unigroup for ₹9,999.33 crore ($1.5 billion) in September 2014. Intel, at the time of the announcement, couched the move as a means to promote the adoption of Intel architecture and communications-based solutions for mobile phones in China and the rest of Asia.
How that partnership has prospered since is not clear.
According to a few industry sources EE Times talked to, the Chinese government’s original plan for YRST was to let the firm take on both flash memory and DRAM production. Over the last several months, YRST insiders have apparently changed their minds. They’re now saying they will only tackle flash memories.
DRAMs, however, remain important to China. Compared to flash, DRAM is where fewer vendors are concentrated for patents and competition.
IC Insights’ Lineback pointed out UMC’s development alliance in DRAM with Fujian Jin Hua Integrated Circuit Co. (JHICC) in China.
This R&D partnership, announced in May 2016, is a part of China’s efforts to expand in memory ICs, said Lineback. “Under the partnership, UMC is setting up a 100-person R&D team in the Southern Taiwan Science Park, while Fujian Jin Hua is funding the technology development.” He added, “UMC does not plan to enter the DRAM market but sees a benefit from developing advanced memory technology with a Chinese partner.”
JHICC has committed ₹37,664.16 crore ($5.65 billion) to build a 300mm wafer fab in Jinjiang City and to develop advanced memory technology and processes. “As usual, the details about the building of a new 300mm memory fab are sketchy,” said Lineback.
Ernst told us that a fundamental barrier for China in the memory market is “an implicit agreement among Samsung, SK Hynix and Micron to keep the Chinese out.”
Some Taiwanese sources offer a different viewpoint, said Ernst. They believe that those who “want to get rid of Samsung’s dominance” might be more willing to work with China. A memory market continuously dominated by Samsung is “a scenario that’s disliked by all downstream users,” Ernst explained. That’s where “China can come to rescue.”
Other Silicon Valley chip executives agreed that they wouldn’t be surprised if Samsung’s competitors turn to China to help slay the Korean goliath.
It’s unclear, though, how much of this is just bluster and pipe dreams.
Even if it succeeds in obtaining core memory technologies, China probably won’t solve its real problem.
Ernst said, “The challenge for China is not merely to catch up with, but to overtake the leaders. Doesn’t this mean a couple of decades of struggle and tons of money?”
This article first appeared on EE Times U.S.