What Will Replace Flash and When?
Will flash memory keep improving into the next decade? IT leaders provide perspective on the great non-volatile beyond as well as an analysis on technology trends, product roadmaps, and more.
Download Dr. Simone Bertolazzi's presentation: What Will Replace Flash and When?
Download Dr. Jeongdong Choe's presentation: What Will Replace Flash and When?
Download Dr. Tom Coughlin's presentation: Why Emerging Persistent Memories Are Necessary
00:00 Jim Handy: Hello, I'm Jim Handy, moderator of the Flash Memory Summit session "C4: Life After Flash." In the next half hour, three highly respected panelists will provide perspective on the non-volatile memory chips of the future. They will give us insight and analysis on technology trends, disruption, product road maps and other memory issues.
Our speakers are Dr. Simone Bertolazzi, a technology and market analyst at Yole Développement. He earned his physics Ph.D. in nanotechnology, publishing more than 10 papers in high-impact scientific journals, and was awarded the prestigious Marie Curie Intra-European Fellowship.
Dr. Jeongdong Choe, who is a senior technical fellow at TechInsights with over 26 years in semiconductor manufacturing and analysis in DRAM, 3D NAND and emerging technologies including 3D XPoint memory. He holds over 100 U.S. patents.
And Dr. Tom Coughlin, president of Coughlin Associates, who covers the digital storage business and technology, a past president of the IEEE-USA with over 35 years' experience. He has earned six patents and written numerous publications, a book, and contributes to Forbes.com and other blogs.
This article is part of
Flash Memory Summit 2020 Sessions From Day One
01:22 JH: Each presenter will speak for five minutes after which we will open the panel to live audience for questions. With that, let's begin. We'll start with Yole's Simone Bertolazzi.
01:45 Dr. Simone Bertolazzi: Hello, this is Simone Bertolazzi, I'm a technology and market analyst at Yole Développement. Yole is a technology and market research company based in Lyon in France, and we have analysts all around the world, covering a broad spectrum of thematics within the semiconductor industry, including memory technologies and related markets, in particular NAND and DRAM, but also a number of different types of emerging non-volatile memories including 3D XPoint or more in general, phase-change memory, resistive RAM, magnetoresistive RAM, and also ferroelectric memories and other types of devices.
I would like to start this presentation for the panel by applying the position of Yole regarding emerging non-volatile memories and their potential for replacing flash in different applications. Of course, when we discuss about this topic, we need to distinguish two very different application spaces, embedded and standalone. In the embedded case, where the memory is integrated within the die in a more complex circuit, there is consensus nowadays in the industry regarding the fact that 28, 22 nanometer could be the last technology node for embedded flash.
02:55 SB: And this is mainly due to economical barriers rather than technical limitations. And therefore, there are significant opportunities for new memories for being adopted and nodes beyond 28, 22 nanometers in particular in system on chips or ASIC type of products or in micro-controllers. And in fact, nowadays we have a number of large players, IBM as well as foundries that are all active in developing new types of embedded memories for nodes of 28, 22 and beyond.
In particular, MRAM is being utilized for its low power consumption in micro-controllers and system on chips and nowadays we have the first products hitting the market. So, in the embedded space, there's definitely a good opportunity for new memories for taking the place of eFlash and nodes beyond 28 and 22. In the standalone business, new memories need to compete with very well-established technologies like 3D NAND for storage type of applications.
04:02 SB: So, at Yole, we do believe that new memories won't replace flash in the next decades, flash will continue its scaling. So recently we have been working on a new report dedicated to equipment and materials for 3D NAND manufacturing, and based on this analysis, we believe that it could be possible for 3D NAND to keep scaling in density and cost beyond 2025 and for more than 400 layers. This will be achieved by string stacking approach, although we are aware that challenges are increasing particular in terms of etching the position as well as metrology.
04:44 SB: New memories in general provide significant technical advantages for certain applications and over existing technologies, but in order to be competitive, they really need to keep improving in terms of density as well as in terms of bit costs that need to be reduced. So, as you can see here, we have different types of emerging non-volatile memory technologies providing advantages in terms for instance of speed and endurance, as in the case of the STT-MRAM technology.
05:17 SB: Although we believe that by 2025, certain memory technologies will not be price and density competitive, where versus existing solution, STT-MRAM, will be approaching the price and density position of NOR flash and therefore can be utilized for certain data and code storage applications. And we'll be starting to take all the NOR flash market, but it will be very challenging for STT-MRAM to be competitive against DRAM or other types of technologies. It will be 3D XPoint that by maintaining an intermediate price and density position between NAND and DRAM, can be utilized for certain applications particularly in data centers such as, for instance, real-time analytics where a large amount of data needs to be stored within the memory and accessed very quickly.
So persistent memory and low latency drives in data centers based on phase-change memory will be the main drivers for the market growth until 2025. In particular, the emerging non-volatile memory market is expected to grow up to four billion by 2025, $4 billion driven by 3D XPoint and low latency and persistent memory applications in particularly thanks to the persistent memory module suite by Intel that are taking off in the data center space.
06:50 SB: I would like to conclude by mentioning that we expect that the emerging memory market, despite is growing, it will remain below 3% of the overall standalone memory market because NAND and DRAM will keep scaling in the future and will maintain the leader, their leading position for a number of applications. And with this, I would like to conclude my presentation, thank you all for your attention and I will be looking . . . I'm looking forward for the discussion. Thank you very much.
07:19 JH: Thanks, Simone. Next, we'll hear from Jeongdong Choe of TechInsights. Jeongdong?
07:30 Dr. Jeongdong Choe: Hello everyone. I'm Jeongdong Choe from TechInsights. TechInsights is the number one reverse engineering company in the world, and we provide the world's largest database on reverse engineering on any field of semiconductor and consumer products. Today, I present briefly the 3D NAND technology and then we will discuss together with you after the panel's presentations.
07:53 JC: 3D NAND flash roadmap on this slide shows all the activities on 3D NAND, so major players like Samsung, Kioxia, Western Digital Company, Micron, Intel, and SK Hynix and YMTC. They are all eager to prepare the next generation on and on. So, for this year and next year, we can see all the 120 layers and 160 something or 176 something on this field. So, you can see the Samsung 128 layers 3D V-NAND and also SK Hynix 128 layers so called the 4D PUC NAND on this slide. So, we just know that SK Hynix and Intel merged together by the end of 2025, so [08:54] a lot of activities are there on 3D NAND. All the 3D NAND buildings and the height you can see from this slide. So especially the Samsung's 128 layer and SK Hynix 128 layer, just to rebuild on the market, and we see that height from source plate to the bit lines already, already reached up to 8 micrometer or more than 8 micrometers. It's amazing.
09:24 JC: So, 3D bit that's a trend, increase, increase on and on, including the SK Hynix, for example 128-layer TLC device. From TLC device, it already reached to more than [09:41] 8 Gb/mm2 but after 150 layers, such as 176 or 240 layers, next year and the following year, probably the TLC with density increased dramatically. From the die density and the number of NAND layers like this one from our expectations, so nowadays you can refer to the logic device, node L7 or L5 nowadays, but stack tiers or decks with QLC devices, we can see that for 128 something layers. But after five years, PLC or OLC or even the stick ties including the TSV such as the HBM-likely structures, you can see there.
10:37 JC: And then after a decade, probably you can see another stack like stack packages PoPoP or something like that. So, we can see the trend of the die size, for example 256 Gb dies from Samsung, 48 layer, 64 layer, 92 layer, 128 layer shrunk and scaled down on and on. About the 500 layers, it's very tiny die will be there, and currently, we can see the 147 total gates such as the PUC 3D NAND from SK Hynix, but after seven or eight years, probably we can see 500 total gates for 3D NAND application. So all the major players shows the different set of architectures such as the V-NAND Samsung, BiCS structure from Kioxia and Western Digital Company, Intel/Micron's CUA floating gate structure and the SK Hynix PBiCS structure, and then PUC structure. And also, we can see the 96 layer QLC die floorplan like this one with the very higher array efficiency due to the CUA structure, but YMTC has the same concept with CUA and PUC. So, peripheral circuits under the array, but recently we found 136 total gates from Samsung and there is also SK Hynix 147 total gates.
12:17 JC: So how about the China NAND? For example, YMTC 3D [12:21] Xtacking applications, the security USB applications, we can find YMTC 32 layers, and 64 layers. And Gloway Pro-SSD applications and YMTC 500 gigabyte and also the one terabyte SSD, you can find the 64 layer Xtacking structure. About the emerging memory? Actually, all the emerging memories, such as STT-MRAM, XPoint memory, PCRAM, RERAM, CBRAM, or FeRAM. A lot of the devices and types are there, but XPoint memory... Especially XPoint memory, we are waiting for the Gen 2, probably they would be a four-stack. But after that, after five years or after 10 years, probably they will need another prototype for the higher density on emerging memory side. So, a lot of activities from STT-MRAM, XPoint memory, and RERAMs are there including all the major players in this area. So, we'll discuss more on this after the panel presentations. Thank you.
13:44 JH: That's really something. Finally, Tom Coughlin will speak after which we will open the floor to questions. Tom.
13:54 Dr. Tom Coughlin: Hello, I'm Tom Coughlin. I'm president of Coughlin Associates. I've been working in the digital storage industry for over 40 years, and in the last 20 years, I've been working as a consultant with my own company, Coughlin Associates. I do various sorts of technical, market, and other types of consulting work, and I also have put on and organized several conferences over the years, including the Storage Visions Conference, the Creative Storage Conference. I put on and organized other events, and I've also been chair, general chair of the Flash Memory Summit for 10 years. And today, I'm glad to be talking at the 2020 Flash Memory Summit about emerging memories, use, why it's growing, and how that's driving capital equipment demand? And this basically follows along another one of the things that I do in my consulting, which reports on digital storage technologies and their applications.
14:47 TC: So emerging persistent memories are necessary for the development of continued scaling of our technologies because flash can't scale with the process advances. Neither NAND flash nor NOR flash, they both are limited at 15 and 28 nanometers respectively. In addition, SRAM scaling, they stop at 14 nanometers. DRAM, another popular volatile memory technology requires continuous refreshes of data, which uses power. So, it's not very good in power restricted environments. But on the other hand, standalone persistent memory applications are growing at the data center, in industry and in consumer devices. And a particular low power, high density memory is needed for embedded applications.
15:34 TC: So let's take a look at some of these emerging memory technologies and how they're being used today. Everspin is a company that makes MRAM chips, they've shipped over a 120 million standalone MRAM chips. They've partnered with GlobalFoundries, which makes 300-millimeter wafers both for Everspin's use as well as for GlobalFoundries' work to develop targeted memory applications with MRAM. And all the major foundries TSMC, Samsung, and many others are starting to ship spin-tunnel torque MRAM products. Phase change memories and other technology is currently in fairly wide use. Intel's Optane technology is available now in NVMe SSDs as well as in DIMM-based products. Those have been shipping in fairly good volumes and increasing volumes over the last couple of years. And both products are now in use in storage systems and tied to the next generation of Intel server technology.
16:32 TC: Embedded MRAM or MRAM in general saves power while reducing chip size and cost. It can shrink smaller than SRAM and flash as shown here, and it also could replace embedded SRAM and NOR since it offers lower power, lower cost and higher density. And you can trade off various characteristics of MRAM for a particular design including data retention, endurance and capacity. Microcontrollers, A6 and other system on chips will lead in the use of persistent memory in embedded applications. Their need for a NOR replacement has driven major foundries, like I mentioned before, to look at using MRAM and also resistive RAM to replace stalled NOR technology as well as MRAM to replace static random access memory. One of TSMC's customers, Ambiq is developing chips that will enable the next generation of battery powered, always on, voice recognition IoT endpoint devices. There a fourth- generation Apollo SoC family sets new standards for ultra-low power intelligent endpoint devices by using MRAM memory.
17:37 TC: The figure shown here is a block diagram of this chip, showing various features that includes, as you can see in the lower left, two megabytes of MRAM. Foundries will drive the use of persistent memory technologies, rendering them attractive, increasing the volume and lowering the total cost of the product, which will eventually replace lower, slower SRAM, and overtime SRAM and upper-level caches. This is from a report that Jim Handy and I recently published called "Emerging Memories Find Their Direction."
The total emerging memory market, MRAM plus 3D XPoint could exceed $36 billion by 2030. This chart shows a petabyte shipment growth that will drive that number. The assumptions behind this chart for MRAM are that it replaces most SoC, NOR and SRAM and has a strong appeal for AI applications. So, we show 3D XPoint in MRAM compare them to NAND and DRAM.
18:34 TC: Development of and manufacturing of MRAM is going to require new capital spending on equipment that is specialized for this for the manufacturer, and we project that by 2030, there could be $700 million in additional semiconductor equipment that would be spent in order to make these MRAM products. So, much of the information in this presentation is drawn from our report on emerging memories, and it's available for purchase. It describes the entire emerging memory ecosystem, the technologies, companies, markets, support requirements and includes embedded memory and discrete memory. 201 pages with 31 tables and 142 figures, and you could see the URLs at the bottom of the slide to learn more. So, with that, I'd like to thank you very much and we look forward to discussion.