Gurtej Sandhu, a senior fellow and director at memory chip maker Micron Technology, is exploring whether DNA can be a storage medium. You read that right. Sandhu believes that DNA storage, or Nucleic Acid Memory (NAM, as he calls it) could be fast and long-lasting.
This is something out of science fiction. Or maybe it’s so wacky that science fiction writers haven’t even thought of it yet. But as Moore’s Law slows down, multibillion dollar companies such as Boise, Idaho-based Micron are exploring ideas for new kinds of memory.
Only about 4 percent of your DNA is actually programming your body. While DNA storage is promising in terms of its ability to retain data, it would have to have some kind of error correction because of mutations. Still, DNA can last from thousands to millions of years, Sandhu said. And it can store archival data for the world using far less space and energy. In other words, NAM could be used as a time capsule for massive amounts of records. Researchers so far have used DNA to store books, songs, and videos in storage devices the size of a sugar cube. Sandhu first wrote about the technology with his colleagues in Nature Materials in 2016.
It’s not clear what form these NAM devices would take, or if we would even store them in our own bodies. Real-world applications range from hard drives to synthetic biology. It could be used to watermark and track genetic content or provide next-generation encryption tools. But the ethics of DNA memory, which has been explored in everything from the Assassin’s Creed video game to Frank Herbert’s Dune novels, must also be explored.
I talked with Sandhu about DNA memory. Here’s an edited transcript of our interview.
VentureBeat: How did you first get interested in storing memory in DNA?
Gurtej Sandhu: In general, I’ve been keeping track of the DNA side of things. The first genome was mapped a few years ago, and there’s been very rapid improvement in our ability to work with DNA, map it and so forth. I ended up in a series of workshops and conferences at Harvard a few years ago. One of the topics was the potential for DNA to store information. George Church was really famous in that space. Not in memory specifically, but he’s been working with DNA since he was in grad school.
We got to talking. He’s a chemist and I’m a physicist, and I just asked him this question. There’s a lot of anecdotal evidence from DNA that’s been preserved — DNA from dinosaurs in Colorado and other places — which people have translated to mean that DNA can stay intact for a long time. I was trying to ask the physics question. We’ve worked with polymers and other chemicals in our industry to try to make memory out of it, and then we correlate that to something called activation energy, and it hasn’t really panned out.
But DNA is a similar material in the sense of it’s a very complex polymer. So I asked him a fundamental question: in physics this is this many electron volts, what’s the activation energy? He gave me the number in chemistry lingo, and both of us were stumped, because we had to translate each other’s numbers. That’s when we said, “Let’s do a study. Let’s look into the fundamentals rather than anecdotal evidence.” That’s what resulted in the paper published in Nature, with Victor. He does the theoretical calculations. I knew about the work at BSU. They’re doing some work on DNA for nano-patterning and other applications. I called them up, and they came in from the DNA side of things.
It’s a long answer to your question, but that’s what piqued my interest. There’s a lot of claims out there, and the physics, the fundamental science backs that up. The interesting thing from that paper was, if you have DNA under controlled conditions — and some of those conditions don’t have to be at the North Pole, so long as you don’t have a lot of moisture — there are practical conditions under which DNA can stay intact for a long time. We started with that curiosity.
VentureBeat: Can you visualize this for me? What would this look like? We know what hard drives and memory chips look like, but what would this look like outside of a laboratory?
Sandhu: That’s the billion-dollar question, or trillion-dollar. Our intent was just to look at the medium, DNA as a medium. With hard drives, we asked the same question about magnetic media a long time ago. Optical media, we had to convince ourselves this was the right way to store information. If the physics doesn’t work, everything else is just hoopla. Now that you’ve established that the fundamental physics is on your side, you come to your question. How do you make it practical?
It’s very different from anything we know of. But in some ways, hard drives were very different from tape drives, which people used before that, and magnetic core memory even before that. Everything that comes along looks quite different. That’s the question, though. How do you make a working system out of this? You have to be able to manipulate this medium, write information to it, as well as read information from it.
That’s still to be determined. There’s some hope in the sense that, in the last 20 years, our ability to work with DNA has come in from the dark ages. But in modern lingo, they may have come in from 2000 BC to 500 years ago. Based on what you need, the gap in performance is still out there. Orders of magnitude improvement are needed in cost and performance to take it from where it’s at today to making a viable memory system. You have to compare cost and performance to what we have today in terms of hard drives and chips.
VentureBeat: Can you explain what would be a bit or a cell in this system? How would you store something in a piece of DNA?
Sandhu: In a digital space, we store information as a zero or a one. In DRAM you have a charge or no charge. In NAND it’s the same way. Each bit is represented by a unit of information storage, a zero or a one, and we call it a binary system. If you have two of those bits — 01, or 10 — so that’s four pieces of information. Two to the power of two. If you have four bits, then it’s two to the power of four, or sixteen different states.
DNA is pretty interesting. It has these four bases. In normal DNA that we have, that turns into a base of four, rather than two. Think about four to the power of one, two, three, four. Imagine four to the fourth power. That’s a lot of information you can store if you have four strands of DNA and each one can represent a combination of four different states.
That speaks well for the information density. In a given amount of material–right now each transistor stores information, each device. In magnetics it’s each magnetic domain down to a certain size. In DNA each molecule stores that information, so the density of information is pretty high. However, how will it actually work? That’s still going to take a lot of work. How do you access and manipulate that information? It’s not going to look like a hard drive. It’s going to look quite different.
VentureBeat: If you’re going to store information in DNA, is it going to be part of a living organism? Or is it going to be some kind of electronic organism?
Sandhu: That’s a relative term. People have developed a lot of synthetic DNA these days. It’s a molecule that looks like DNA. Even in nature, if you think back, nature probably went through a lot of designer DNA that eventually led to life as we know it. Even the DNA we have in our bodies, how much of that is support material and how much of it is actually the program of life?
But to completely stay away from that controversy, there’s a lot of synthetic DNA. These are organic molecules similar in function to DNA, but they don’t mimic any life form that we know of, or anything that can sustain life. That may be a way to do a way to do this. We’d have to make large quantities of DNA, and the DNA that’s in our bodies was nature’s miracle after millions of years. That may not be the lowest-cost DNA for this application. There may be designer DNA molecules we’ll build in a chemistry lab and use, and those will not necessarily be associated with life as we know it.
VentureBeat: This did make we wonder if we could just store information in our bodies, though.
Sandhu: If you want to go back 15,000 years, in the Indian Vedas — I’ve been reading up on this lately, so I’ll bore you for just a moment — those speak of our body having eight different types of our memory. The memory we store in our brains, in our mind, what we pick up as we learn over time, is only one of those eight. Their point was, our body–for example, the skin, the features on your face, that’s actually a memory of your forefathers, for however many generations you want to go back. Our body stores a lot of information, but over a period of time. We call it evolution, the features we pick up as the generations go by. So in a way what you’re saying is true, and even more than you might realize.
VentureBeat: There’s a famous video game called Assassin’s Creed. The premise is that you can go back in time through your ancestor’s DNA and relive their memories, their lives.
Sandhu: Yeah, I’ve seen my kids play it. Even modern science backs this sort of thing up now. Only four percent of your DNA actually seems to be programming your body. The rest of the DNA just sits under that four percent. But we don’t know how to manipulate this information. What we can manipulate and learn is the information in our brains. That’s not necessarily DNA storage, but long-term memory, over evolution, over generations, that could be a similar concept.
What looks like science fiction is sometimes borne out once we learn more about the science. Sometimes it’s only fiction, and sometimes–the Vedas also talk about 31 dimensions of the universe, and string theory is up to 13. They’ve solved equations up to 13. Is it going to reach 31 eventually?
VentureBeat: How have people reacted to this, whether your colleagues at Micron or other scientists?
Sandhu: DNA always appeals to our imagination. It’s such an important part of us. The human body is the most sophisticated and complicated machine ever built, whoever built it. None of our computers, in complexity and capability, come close. Our brain processes a humongous number of transactions every second, based on our best estimates. Since DNA is part of us, we should appreciate it, I suppose.
In the technology side of things, we’ve tried to work with polymers and so forth, which technically speaking are similar molecules. The fundamental bonds are hydrocarbons in both cases. Hydrocarbon bonds are very difficult to work with, because they’re not very stable. But here is this DNA material, which has very similar fundamental bonds–this is a big part of why we did the study, because I was very intrigued. How can the same hydrocarbon bonds, in the form of a DNA molecule, turn out to be so stable?
What comes out is, your DNA–as you know, it folds and unfolds as part of the process of normal life. It has to work within a certain temperature range. We’re pretty fragile in that sense, with a narrow temperature range. But DNA folds and unfolds and can change the stability of the molecule based on its configuration. In your body, DNA has to be easily manipulated when cell division is going on. It has to unfold and split into two very easily. But then, once that cell turns into your skin or whatever, it has to be very stable again.
So DNA is able to perform functions that are pretty amazing. It can be both unstable and stable based on its configuration, even though the fundamental hydrocarbon bonds are the same. Our study showed that all this DNA can be stable for hundreds of years, even though it can also perform millions of cell divisions in our bodies each day.
VentureBeat: We’re all familiar with evolution and mutation. Does that play in to what you’re working toward? Can it help you or hinder you?
Sandhu: Before I answer that question, one other thing I found very intriguing–your body has so many cell divisions going on every time. There’s a natural self-repair mechanism going on in that process. With millions of divisions going on, and environmental factors — light, cosmic rays, all these things — a very high percentage of these events have an error. The split doesn’t go as planned. There are some estimates that, based on our error rate, our bodies should simply disintegrate in less than 24 hours.
Nature has something called self-repair in our bodies, working in the background. Every time something goes wrong, your body corrects for it or compensates for it. That piece is pretty important. So, going back to your question on mutation, that happens over generations. As people have shown through the study of evolution, it happens in reaction to your environment. That’s a long-term process that happens for survival. If you move from a tropical place to someplace cooler or vice versa, over time people with darker skin are selected, or something like that. Mutations are favored and people start looking in a certain way.
The short-term errors I talked about and then longer-term mutations, you could say they’re linked in some ways. Whatever survives the most in a given environment over time, that’s evolution, or a selection process. But it’s the same fundamental mechanism in the DNA that drives both. In the case of memory systems using DNA, there’s also a notion of self-repair and error correction methods that we can apply to keep our information intact.
VentureBeat: Our bodies have error correction, then.
Sandhu: They do, or self-repair. Self-repair is an active process, which I found very amazing. Error correction is just something where you say, “Okay, this is wrong, I reject this piece of information.” Which is important, but to be able to find an error and repair that–we have ways to repair as well, in a consistent way. That’s pretty interesting. Our body ends up repairing itself.
VentureBeat: Why don’t I get fixed to the point where I look like Brad Pitt, then?
Sandhu: [laughs] Nature still does its own thing. Everybody being Brad Pitt might not be the best outcome for humanity.
VentureBeat: How do you proceed from here? How much support do you need for this as far as grants or other ways of moving forward?
Sandhu: There’s always excitement any time you discover something new. We have a new group of people working on this. But from our perspective we have to look at the larger piece around how this will become practical, and what might be the challenges in getting there. Once we identify the challenges, we’ll try to encourage research and work in that space so we can solve some of these problems.
My goal is to make sure the physics is right. Once I’m convinced of that, there’s a potential in this medium. Now the question is, how do we make it real? There are all sorts of claims, but from our business, we know how memory systems work, and we look back at the history a little bit. We participate in the SRC consortium, which is funding some of this work. IARPA is involved in funding one of the projects in conjunction with SRC. IARPA, as you know, are the same guys who invented the internet, funding research into that.
A couple of years after we published our paper, we’re seeing other companies — Microsoft and others — a lot of people paying attention to this. Once you have that type of activity going on, hopefully the focus shifts into making this real. Cost, performance, making the system work. The starting point of this is archival memory. That’s where you have hard disks, or tape drives. It turns out that tape drives have to be rewritten every five or 10 years. It’s not very persistent memory. With the type of information we’re generating these days, rewriting those things is a big part of their cost now, if you look at it over 100 years and beyond.
We’re producing so much information now compared to before, and we want to store it forever. In order to store it forever, you have to rewrite it, and that cost is significant in terms of energy. That’s one of the interesting possibilities for DNA. If something can stay stable for 100 years, as opposed to five years, so you don’t have expend energy maintaining it–the other part, if you run the numbers, a teaspoon of DNA could contain all the information we know today about our universe. It can hold that much information. The form factors are pretty attractive. The challenge is just how you read and write into it.
That’s where the focus is now. You have to design systems and come up with a way to read, write, and manage this information, and then target specific applications of interest. That’s why we’re helping direct this research funding. The IARPA project I mentioned, which we’re part of, and SRC as well, is right on the money. It’s focused right on those things. It’s been good to see the synergies we have. There are opportunities to expand further from there, too. In a way this is just the start.
The problem, as I say, is pretty basic. We’re creating so much information, and storing that over long periods of time is a challenge. We see that getting exponentially worse over time.
VentureBeat: How much activity like this is going on at Micron? Is it a large department, or fairly small?
Sandhu: Looking at different types of memory for the future is a pretty big activity. I was doing DRAM in the ‘90s, and in the early 2000s I went to NAND, when the iPod happened and the whole NAND thing exploded. I was on the early side of those things. We’ve had a find ways to get ahead. Then, in 2006 to 2008, I started this whole–what comes after NAND? What else is out there? We’ve looked at many different applications, many other types of memory — spin torque RAM, phase change, all kinds. We looked at every mechanism out there we know. Each time we looked at the physics of it, and then what kind of system we could build, and what would be the end market.
Out of that activity, DNA is the latest one we’ve identified. From those previous activities, there are some new memories we’ve put in the market. You’ve probably heard of 3D Xpoint under the different names we call it. DNA is now the latest thing, and it’s potentially an archival memory space. But it really challenges your imagination. It could go so many different ways if we could master the engineering side of things.
It’s a paradigm shift any way you look at it, but there have been similar shifts in the past. They look normal now because we read about them in books, but the background behind it, if you consider what it felt like at the time, it was pretty revolutionary. The whole silicon chip revolution, we went so many directions before we got here.
This is interdisciplinary in many ways, even more so than it was in the past. The disk drive was magnetics and then micromechanical systems. Those were different fields of engineering and science. This one challenges us even further. It goes into chemistry and biochemistry, and maybe even health sciences in some ways. On the other side, you still have to get the information into a computer. The chip side of things, the electronics side of things, they all have to be considered in the system. How do you merge this in a way that it all talks to each other and you can build a system out of it?
VentureBeat: I’m curious what you think of some of the other experimental branches of storage research.
Sandhu: The one with the most history and activity and traction is optical and holographic. That’s been worked on for a long time, obviously. You can go out and buy an optical drive. One thing I like to think about–I always go back to physics. I like to start there. If you think about a photon, a green light or a blue light, that’s just the smallest unit of information you can store, the least amount of energy you can have. With a green light, the size of the photon is 5,000 angstroms, 5,000 nanometers. Comparing to that, a NAND device today is about 20 nanometers, and getting smaller as we go. From there, a DNA molecule is just a couple of nanometers.
The issue with optical-based systems is that the light wavelengths we work with have to be in the optical spectrum. They’re relatively large. You have a disadvantage where the fundamental unit of information is pretty large, and then you try to compensate for that by using other engineering means. Holographic memory is one of those things. You try to go to three dimensions and pack in more information. But the fundamental limitation is there.
That’s why people don’t talk much about optical drives to replace tapes. People use them, but they haven’t taken off in big way. Part of that is the form factor. These are bulky devices, because the fundamental unit of information is pretty large. That’s where the challenge is.
VentureBeat: Do we foresee the same kind of wall that semiconductors are expected to run into?
Sandhu: We have Moore’s Law, and that’s been an interesting challenge for the last 30 years we’ve been driving it. Now some people worried about how we’re in the “more than Moore” space, which is true in some ways. Planar NAND devices stopped shrinking. We’re into what’s called 3D NAND now. We use three-dimensional space and stack them up. That’s certainly true.
The way I put it is the simple days are over, where you’re shrinking things 30 or 50 percent every other year and seeing if the device works once you shrink it. As an engineering exercise that’s great, and you want to keep doing it until you get to a dimension where the physics you’re using don’t work anymore. Or the physics are okay, but the engineering is cost-prohibitive. The gods of physics don’t want to help you. It’s always physics that limit you when you’re moving down the path of technology, either by simply not working because it’s not feasible, or by making it too expensive to get more than tiny benefits. That’s where we’re heading now with Moore’s Law.
You can’t imagine life without NAND or DRAM or these systems we have today. About 95 percent of the devices built all over the world today are memory devices. People don’t realize that. I have a slide that I didn’t believe for a while, but we’ve run the numbers again and again. We hear about microprocessors and foundries and all that’s great, but the number of devices that go into those chips, if you count them all up, 95 to 99 percent of them are all memory devices, storage-related devices. It’s a huge thing for us. We can’t just not have them.
We have to find ways now to sustain technology progress, which means we have to be more innovative. You can’t just be doing the same Moore’s Law shrink before. We have to go to the circuit level, to the chip level, to the system level, how you put the chips together. There’s always room for innovation. It’s just that sometimes we put the burden for innovation on only one thing — keep shrinking the devices.
From that perspective, it’s going to be so much more interesting going forward in terms of the innovations we can put in a three-dimensional stack. There’s always room to innovate, and you can get a lot of ROI out of that. You just need a different paradigm. The whole ecosystem has to be able to accept it. There’s room for it and need for it and a lot of market pull to enable those systems in the future. It’ll keep sustaining our progress.
If you think about NAND, if there were no NAND, there would be no cell phones. That’s a pretty simple statement, but people don’t realize it. It’s the truth. The phone you have in your pocket wouldn’t happen without NAND, and that technology, I was working on it in 2003. It’s 15 years old.
VentureBeat: Hopefully you have less than 15 years to go on this one.
Sandhu: [laughs] There are lots of things like AI and machine learning at work. It’s an exponential curve of our ability to use information. That’s very important now. About 90 percent of the information we used to store a few years ago, it would just sit there with nobody using it. The example I use now is a picture on Facebook. A few years ago it would just sit there, but now there are algorithms that can go through and recognize your face and flag it. It’ll automatically do that. These little algorithms and capabilities can use and enhance the information you’re storing. AI and machine learning are going to do huge things in that space. Not only are you going to store a lot of information, but that information has a lot more value.
Today we have a $100 billion DRAM industry, and approaching $100 billion for NAND. Just a few years ago, maybe 10 percent of stored information was actually used for something. Imagine if you can use all of it for some purpose, commercial or enhancing standard of living or something. Then the demand picks up even further.
VentureBeat: Is there anything else we’ve neglected to talk about so far?
Sandhu: One thing that I hope can happen when people take a look at this–no amount of technology progress happens without young minds getting into it, getting interested in it. That goes back and forth. Fifteen years ago “nano” was a big buzzword. Nanotechnology came into the press and there was a lot of excitement about that. The good thing about that is it gets into the mainstream of culture, kids start paying attention to it, they go to college, and we get a good influx of people, young engineers who are excited about this.
Without those people, we can’t have progress. Any time we can have something in the press that challenges the imagine and gets young people to become interested, we always welcome that. It brings us a new infusion of smart minds. When I talk to students I tell them, “15 years ago this phone wasn’t possible. It’s possible because of this chip.” People take things for granted. Getting young minds to feel that this an interesting field where they should pay attention to how it benefits them, and mankind in general, that’s important.
VentureBeat: We’re going to need a memory system for all this artificial intelligence everyone is looking forward to.
Sandhu: If you merge AI, machine learning, and DNA together, that catches people’s imagination, right? We’ll have machines coming with DNA in them. [laughs] That’s not the right way to say it. It’s not true. But it connects with people.