4DS Memory (ASX:4DS): Interview with Chairman Wilbert van den Hoek
November 15, 2021
4DS, 4DS Memory, ReRAM, video
4DS Memory (ASX:4DS): We spoke with Chairman Wilbert van den Hoek about what caused 4DS’ recent trading suspension and the development pathway in the near term.
See full transcript below.
4DS Memory will be presenting at the
Stocks Down Under Semiconductor Conference on 30 November 2021!
Marc: Hello and welcome to “Stocks Down Under”. Today we’re joined by Wilbert van den Hoek who is the Chairman of 4DS Memory, and David McAuliffe, Executive Director. Welcome, gentlemen.
David: Hi, Marc.
Marc: Wilbert, you joined 4DS I think about a year ago. I looked at your profile and it’s quite impressive. So when I started out as an analyst I covered Philips semiconductors, it was one of them, and I see you’ve worked there as well on the research side. Maybe to kick things off, can you tell us a little bit about your background in the semiconductor industry?
Wilbert: Yeah. As you mentioned, I started my career joining Philips Research. Worked there for 11 years. Both in the Netherlands, in the Central Research Lab, and then the satellite lab in California. And after 11 years I wanted to explore what was going on in the semiconductor industry in Japan. And Philips wanted me back in the Netherlands, so we parted ways, and I joined Novellus Systems Semiconductor Equipment Company, to set up their direct operations in Japan in the early ’90s when Japan was ruling the semiconductor world.
Since then, I returned back to California and I’ve been in California ever since. And stayed with Novellus for a total of about 18 years. And the last role I had, I had a series of executive positions in the company expanding from running their R&D organization, developing a new product, taking it to market, running a business. I was responsible for manufacturing, I was responsible for the human resource department, and I was acting general counsel of the company. So very broad experience base.
And my last role was running the Novellus Venture Fund. They had a small strategic venture fund and that introduced me to the startup world. And so since 2008, I’ve been active as a board member adviser to start-ups, as well as large public companies. Probably the most famous one is, I was involved with Cypress for I think about 12 to 14 years, half of that as a technical adviser and the last seven years as a board member.
Marc: Right. And that was acquired by Infineon in Germany.
Wilbert: I left in 2017. And then two years later, the company was acquired by Infineon.
Marc: Okay. Well, it sounds like the right guy in the right spot with, you know, obviously, you know, with your startup experience at Novellus, with 4DS, obviously, still in that stage basically. What attracted you to the company when you joined last year?
Wilbert: Their value proposition. As you may have seen, I’ve recently written a few papers with professors at Stanford Georgia Tech, on what is going on in the memory world, and what things like IoT require. That was focused on the embedded memory space but there’s also a tremendous need for innovation in the standalone memory space in the memory used in microprocessors for desktop, laptops and server applications, and 4DS has a unique offering in that space. And so I was interested in helping bring that to market.
Marc: Yeah. So I just wanted to touch on, sort of, you know, more recent developments, obviously, with the trading halt or the suspension recently. Can you talk us through why that suspension took three weeks to get through and what was behind it?
Wilbert: Before I do that, let me explain a little bit the space that 4DS is in. Because, I mean, we’ve gotten feedback from shareholders and there seems to be some confusion. Like I said, the memory space is broken up into two areas, memory used in standard computing and what I would call embedded memory, that is very important if you’re working in the IoT space for microcontrollers. 4DS memory is focused on the memory used in computing. And in computing, there are, high-level, three types of memories.
There is the Static RAM, which is embedded in the processor, typically about 75%, 50% to 70% of the size of the chip, or the area of the chip is consumed by Static RAM. That’s where all the computations occur. Then on the motherboard next to the microprocessor, there is a DRAM. That is the memory that’s actively accessed when the computing is performs. It’s also the memory, for example, that hold the operating system. And then there is the memory that is used for storage NAND, typically. Used to be hard drives and these days it’s solid-state drives. And that is the NAND memory.
These three types of memories have very different characteristics. NAND is what’s called non-volatile. That means that whether you pull the plug of the power the memory stays. And the goal of NAND is to be able to store data, pictures, programs for 10 years. You have the SRAM, that is purely non-volatile. That means that if you turn off the power on your computer whatever was in that memory is gone. And then DRAM is also non-volatile, but even if the power is on the data disappears if you don’t do anything.
So the DRAM is non-volatile but every 60 milliseconds or so you have to rewrite that data into memory because the charge leaks away. And 4DS is aiming to replace and augment the DRAM. So very different from other types of memory. Obtain is another memory that Intel and Micron have introduced in that space. Obtain is very close to NAND memory. So it’s targeted to be non-volatile, to store data for 10 years. It’s targeted to be faster than NAND, and in server farms, the Obtain replaces part of the NAND memory but so they’re very close to the NAND side. 4DS is very close to the DRAM side.
So that’s high level where we as a company are positioned, standalone memory. So we target to build chips that contain a memory and they go on a board. And we’re looking to introduce memory that is high speed as needed to replace DRAM. And it is non-volatile. We call it persistent, the industry calls it persistent. So it’s not a memory that will store data for 10 years, but it stores data much longer than a DRAM. It could be hours, it could be days, it could be weeks, it could be months, but definitely not targeting the 10 years.
And one thing that the industry has struggled with over the last 25 years, is to find replacements for these three core types of memory SRAM, DRAM, and NAND. And for a long time, people were working on what’s called universal memories. A new technology that could do everything. That have the speed of an SRAM, would have the storage capability of a NAND. And the industry has been unable to do that because there is a trade-off that you have to make. And so 4DS is targeting to make the trade-off to make memories that are more like DRAMs. They’re non-volatile, but not permanent like a NAND.
Other people, and if you look, for example, at MRAM technology. MRAM is a new magnetic memory technology and they’re targeting solutions that are like a NAND and have 10-year storage capability. So there’s this wide space of choices you can make in memory. And at this point, 4DS’s goal is to offer the industry trade-offs. Is how much endurance…? Endurance means how often you can write a memory, how much do you trade that off with retention? That’s how long data stays in the memory before it spontaneously disappears. And so that’s been the focus of the company.
Another thing that’s important to understand in the status where 4DS is. When you build memory, you first build the memory cell. So you make a new type of memory and a DRAM as a capacitor. In an MRAM it’s a magnetic switch. In a NAND device, it’s a transistor with a floating gate, either a Polish silicon floating gate like Intel or Sonos floating gate like everybody else in the NAND space. So they have different types of characteristics to store the data. And 4Ds has a resistive RAM technology.
So we have focused on developing our memory cell and get the right characteristics of the memory cell. And we reported in August that we had made a dramatic improvement in the performance of this memory cell. We also reported in August that our attempt to put that cell into a memory array, where the memory cell is driven by a transistor, had failed because of misprocessing. And we had, of course, after the misprocessing you need to go analyze what went wrong. We had extensive discussions with IMEC, which is our foundry partner, and we had identified our paths forward, and we had communicated that in August.
That we were gonna switch to a different platform. We were gonna have a one-megabit array, and that was the path forward. And we said we are finalizing the date at which that memory platform we lost can start. And we had said either it’s gonna start in late September or in October. And we had to do some pre-work to determine what was the right time to start. And that work was occurring between August and the beginning of October. By I think, October 8th we believed we had all the information.
We were digesting it internally and the company went into a trading hold for a few days to make sure we had all our ducks lined up, and we were scheduled to announce, I think on October 11, what our plan forward was. Literally, hours before we were releasing the press release, we got a message from our partner in our regular Monday morning meeting, Monday morning California time, that they had provided us misinformation. That they had looked into what they told us turned out to be incorrect relative to this new platform.
And the reasons why the prior device lot had failed, would also be the reason why this new device lot would fail. Because there was something in the design that they thought wasn’t there, and it turned out to be the same design flaw in those two types of devices. So for three weeks after that we have been negotiating with IMEC, what can we do to make sure that device problem is fixed? So that we have a high probability that the platform lots come out properly. That caused about a three-month delay in our timeline that pushed all the processing into 2022.
That required us to negotiate an extension to our collaboration agreement with IMEC to make sure that we didn’t forecast a plan that was not supported by a contract so that IMEC would execute their part of the activity, and that took three weeks. And I believe actually that those negotiations were done at high speed and high pressure in order to minimize the time but it still took three weeks before we had a resolution to all these issues, and we could announce the press release we made recently, where we said we have everything cleared up, we are instead of simply starting the device lot, we’re gonna do a short loop first.
Do some characterization to make sure that things we are concerned about in the processing will not cause the lot to be defective. And then we’ll start the device lot early next year and expect the device lot to come out around the July timeframe.
Marc: Right. So…
Wilbert: So a long explanation for what turned out to be a pretty painful but relatively short in light of all the things that needed to be done, a trading hold for the company.
Marc: Yeah. Okay. Obviously, the market was extremely negative on the news, obviously. So I think in part the issue might be for the market that it’s uncertain as to, you know, the delay, the actual delay. You get three months, but maybe the market is assuming it might take a bit longer. But in any case, it scared the market, obviously. But from a development point of view and timeline point of view, what needs to happen, say right now and in the next couple of weeks, before you can get that, sort of, new lot started?
Wilbert: We’ve identified the design issues and we’ve identified unit process challenges that need to be addressed. So the process has started at the new design, the new mask is being designed. That’s a fairly long process and we have started a short loop where we’re building a way for…with just a memory stack so we can optimize the edge process that was part of the root cause of the failure of the platform lot, I think, we called it the second platform lot results we announced in August.
So, instead of starting the actual device lots that contains the megabit memory array, which was in August announced to be occurring in the September, October time frame, this will now start sometime in Q1 of next year after we have the results of the short loop. The short loop will be processed early in Q1. Exact timings are not being communicated because what probably will happen in real life, the lot will be started before we have the results on the short loop the device lot will be started. It would then be put on hold for a period of time to digest the results of the short loop, and then it will be continued processing.
And that’s why we announced there’s a three-month delay. In August timeframe, we were expecting the device lot, the platform lot to come out in April and at this point, our best estimate is that the device lots come out in July hence the three months delay.
Marc: And, I guess, in terms of restoration of shareholder value, maybe this can be a question for you as well David in terms of, you know, rebuilding that confidence and the value. So you think this process that you’ve now laid out, should restore that shareholder value by that point?
David: I do. I think it’s been a very well-designed plan going forward and, you know, as always with these things it’s never a straight line, we’re always going to have hiccups in a development company like ours. Unfortunately, I think it came as a bit of a surprise to us as well. We’d said we thought we’d be making stuff in November. But you have to deal with the situation at hand, it’s something that was out of our control. I think the team in Silicon Valley did a very good job working with IMEC to come up with a sensible plan which gives us the best chance for a third platform lot coming out in July to actually give us what we want ultimately.
And during the course of that for between now and then, I’m sure we can restore shareholder value as we lead up to the results of that work that needs to be completed. That tends to always happen with our organization, it seems to be that, you know, as we get closer to coming out with the results the value of the company significantly increases. And I think that it’s very likely will be the case here.
Marc: All right. And just in terms of the strategy or the exit strategy, I should say, previously it’s been communicated that you’re looking for a strategic buyer to just basically, instead of licensing it out to various partners, to sell the technology. So basically, two questions. First of all, is there any change to that strategy or can you update us on that? And is the timeline set for mid-2022, is that also the relevant timeline to be looking at that more closely, more specifically?
Wilbert: Yeah. That strategy is still the right strategy for the company. When you’re in the standalone memory space, and memory is very cost-sensitive, price-sensitive, the only way the history has shown you can be successful if you have a dedicated fab for that memory. So today, if you look at DRAM, if you look at NAND those are not made in foundries, they’re made in stand-alone dedicated factories that only do that type of technology. And those fabs as a minimum, cost $5 billion to start.
So early on 4DS has recognized that a strategy where the company was going to raise that kind of money to build their own factory was not a practical strategy. And so, what we are doing is, we’re doing the R&D work for the big memory guys. We show them a working technology and at that point, they can adopt that technology, and they will have to compare our solution to solutions they’re working on internally in their own R&D facilities. So yes, that strategy is still the same.
In order to convince a memory company to buy technology from a company like 4DS, there is a minimum data set you need to demonstrate, and I think the company has communicated that clearly over the last years. We’ve always talked about platform lots, platform lots that have megabit they’re raising them, and demonstrating a functional megabit array is a critical step in being able to convince a memory company to compare its internal development to the solution that a company like 4DS is offering.
And so, what we’ve been working on and what this next platform lot is to produce that megabit array, and when that is successful, we’ll have a critical data set that would allow the start of an exit process. How long that process will take, that depends. It depends on how convincing our solution is, that depends on what the internal solutions are for the customers, the market dynamics, but a critical deliverable before we can even consider talking to potential acquirers is provide that memory macro, that memory array data to the people. To quote an unnamed executive of one of our potential acquirers is, “We get an idea for a new memory stack once a week.”
He said, “We’re not interested. We are interested in people who can demonstrate that they have a new memory stack and a memory array.” And so, they’ve solved all those issues because realize, from that point on there’s still probably a two-year timeline for the acquirer to commercialize their product and put a real product on the marketplace. So there’s significant investments still to be done after that and there’s significant risks they have. So they want to have at least that point proven to them so that they can make a proper assessment whether this is a technology that they want or not.
Marc: Right. And so, we’re talking about that’s the megabit array, sort of, roughly mid next year is when you expect to have that available?
Wilbert: Our next platform is aimed at producing that information, and that is critical collateral in order to start a process of finding a buyer for 4DS. Yes.
Marc: All right. Excellent. Well, thank you very much for your time, Wilbert and David.
David: Thanks, Marc.
Marc: We will obviously keep a very close eye on the developments because it’s exciting times. Thanks for your time, and yeah, hope to speak to you soon in the future.
Wilbert: All right.
David: Thanks a lot, Marc.
Wilbert: Nice talking to you, Marc.