Paul: There’s never been a time when a resource has held as much power to shape our world as data does today. I’m Paul, and on this podcast, we explore how innovators are using data to transform how we live, work, and create. Today I’m joined by Stephanie Hernandez, senior engineering director at Seagate, to discuss Mozaic, a breakthrough technology that arrives at a critical moment in time when AI is generating unprecedented levels of storage demand. Let’s get into it. Welcome to The Data Movement.

Stephanie, welcome to the show, first of all.

Stephanie: Thank you for having me.

Paul: Yeah, of course. Amazing to have you here. I’m super excited for a variety of different reasons to have this conversation. I want to start by going back in time. You’ve been with Seagate for 15 years, and I want to explore a little bit around your journey with Seagate, your perspective on the storage industry and the shifts that have happened in that time. It feels like we’re in the midst of one at the moment. So, let’s maybe let’s start there. What have you been doing at Seagate for the last decade and a half?

Stephanie: I started at Seagate as a reader designer. You know, going back a little bit further, I did my PhD at the U of M, and I actually was working on magnetic recording even as a graduate student. So, I worked in the lab of a professor called Randy Victora, and his focus was on designing and looking at advanced HDD technologies, and his work was more computational, which really fit in with what I was interested in at that time. And you know, he was funded by Seagate, and that of course led to internship opportunities at Seagate and eventually landing a role as a reader designer here in Minnesota about 15 years ago. And I was a modeler. So, meaning I used physics-based models to understand how our reader designs should scale as a function of aerial density and to get the performance that we need. Eventually, I moved on to the research group, which fit a lot better with what I’m interested in. I like figuring out what will be the technologies 10 to 20 years into the future. I joined the Seagate Research Group, also here in Minnesota. And then I really started modeling heat assisted magnetic recording. At that time when I joined, it was a future technology. Now it’s a reality. Now I lead a group that’s looking at the future: HAMR technologies and even looking at what are post-HAMR, HDD and alternative data storage technologies.

Paul: Yeah. Amazing. Let’s start with talking a little bit about HAMR. Super topical at the moment in the industry. Seagate, at least by the time this episode publishes, we would’ve, made a big announcement around new aerial density capabilities that are unlocking capacities into the 40 terabytes per drive.

And some of the things that our customers are doing with those drives today. Before we get into that, I want to go back to 13, 15 years ago when you started. What was the perception of HAMR at that point?

Stephanie: Throughout my Seagate experience, I have experienced that transition from perpendicular magnetic recording to HAMR in the design groups. And, you know, I can’t overstate how significant a transition that was. Before HAMR became what most of us are working on, people were skeptical.

There were still a lot of challenges that needed to be resolved in order to bring that technology to pass. There are significant changes that were required. You needed brand new media design, brand new head design. You needed a head that has optical elements. You still needed reader technology that keeps up.

With the aerial density and capacity increases, all the parts of the recording system have to be designed to be able to support that increased capacity. So, there were still a lot of issues and other challenges that needed to be resolved to get to where we are today. So certainly, there were a lot of skeptics at the time, but you know, as we learn more as the design evolves, it becomes clear that yes, this is fundamentally a technology that’s very viable and the challenges are engineering challenges that through continuing to understand HAMR, we can overcome them.

Paul: And I imagine there was skepticism around whether the technology could fundamentally work, or was it more about at that pointat least, we’d proven that it can work, it’s just, can it scale to the level that the previous Gen PMR technology could?

Stephanie: I think people have always thought that it’s a technology that could fundamentally work in terms of yes, you can record smaller bit sizes with HAMR, but how do we design a system that can operate in real life out there ... in the real world with the real reliability and performance requirements that are necessary to deploy such products?

We always thought it was scalable. I mean, it always became more clear as time went on. But certainly, making sure the devices were robust, that was one of the biggest challenges.

Paul: It’s, so, interesting to me because I’ve got on my desk here a prop, and this is a Seagate drive maybe actually from about ... it’s probably close to 15 years old. So, when you first started looking at this stuff when you joined, it’s a four terabyte 3.5-inch drive and one of the most interesting things for me is about hard drive tech is the form factor is like, even though it’s a 15-year-old product, the form factor, the dimensions of the thing, this little gray box is the same, right? And there are specific reasons for why that is. Like you can’t change the form factor. So, all of the innovation that you just alluded to at a really high level is actually going on inside this little rectangular box, which is just such an interesting engineering and creative challenge. Can you talk through the other thing in relation to that question, is this idea of like superparamagnetic recording ...

Stephanie: Limit.

Paul: Limit, sorry. Yeah, super paramagnetic limit. Can you talk about what that is and why? Why it’s important? I want to dig into the technologies inside that helps solve for that.

Stephanie: Right. So, for many years people have been predicting the death of a hard drive, right? What is the limit to how small of a bit you can record on magnetic disks in a hard drive? And so, the super paramagnetic limit ... it’s just saying if you reduce the particle size of a magnetic material, it becomes very thermally unstable.

So instead of magnetic, it becomes paramagnetic. So, it’s not magnetic anymore. It doesn’t hold any information. And you know, we’ve had many different technologies ... technological evolutions over these 20, 30 years and before then, of course. And each of these critical technological changes have provided an opportunity to keep beating the superparamagneticmagnetic limit.

We’ve been able to incorporate new designs — new media designs specifically — that are able to support smaller and smaller grain sizes. So, the disks, in the hard drive, the recording, material is a magnetic, it’s a granular, magnetic material. And you have these grains, which are right now they’re about 10 nanometers, less than 10 nanometers.and, each technological evolution has allowed us to bring a new media design that is able to support smaller and smaller grain size. So, we’ve been able to beat the super paramagnetic limit, and we don’t know exactly what, when the actual limit will hit us.

I mean, eventually, yes. But we think that technology, the technology that we’re at right now, heat-assisted magnetic recording, can take us pretty far.

Paul: Yeah, and the talk about ... tell me about the grain size and why that’s important in this kind of equation.

Stephanie: Yeah, the information, it’s written on the disk in these tracks. So, tracks are the sequence of bits. And the sequence of bits, determines it; it makes up the data, right, that’s written on the disk. And each bit is a fundamental unit of information. It’s a one or a zero, or one or a minus one. And these bits are collections of grains that are magnetized in the same direction. So, in order to keep reducing the bit size and maintain the signal-to-noise ratio. So, the signal to noise ratio is a really fundamental, metric that’s important to us, because we want to sense a strong signal from the media with reduced noise as much as possible. In order to reduce the noise, we need more grains within a bit.

So, I can’t just reduce the bit size, without reducing the grain size as well, so that’s why it’s been so critical to shift to different recording technologies that are able to support smaller grain size.

Paul: You alluded to a few of the innovations that make up HAMR or Mozaic as we refer to it as at Seagate. What are some of those kinds of fundamental components or subsystems that make that up?

Stephanie: Yeah, I talked a lot about the media, so you know, maybe I’ll start there.

Paul: Yeah.

Stephanie: For HAMR, the media’s fundamentally different than PMR — completely new media design — the material’s iron platinum based, and it has very high magnetic and isotropy. That means that I can, it’s always challenging to fabricate media with small grain size and very good properties, but by virtue of having that high in isotropy, you could push the grain size much further than you can conventional media designs.

You know the reason we can use iron platinum is because now we have a brand-new writer. In perpendicular magnetic recording, we have magnetic writers that supply a field. That’s if this is the media plane, then the field is perpendicular to the media plane and the bits are magnetized, also perpendicularly to that media plane. 

With HAMR, that’s all still true, but now we need an additional excitation in order to write this very high isotropy media because iron platinum is very, thermally stable. Very, magnetically hard, but that means it’s very hard to write. I can’t just take a PMR writer and record information on it because of that robustness. So that means I need to supply some kind of assist in order to be able to record information on this new high, isotropy media. And the best way — really the only way — we could really do it is by applying heat. But you want apply heat, not all the time, but only during the right process.

Magnetic materials have this property where they lose their magnetization, they lose their isotropy and magnetic hardness as a function of temperature. We want to apply just enough temperature, but only when we want to write the bits. And that’s why a brand-new wri needed to be designed, that has a magnetic writer because we still need that to supply the excitation needed to write the bits on the disk.

But we also need now an optical writer, which is brand new to HAMR. You need a laser. That applies that heat input. You need an optical wave guide that will take that energy from the laser all the way to the air bearing surface, which is that area right next to the media. And then, it’s an innovation called a near field transducer that can efficiently direct that energy to apply a very narrow concentrated heat pulse on the media.

So very different media, very different writer. Right? Those are the, let’s say the fundamental changes, however, the entire system has to scale. You need readers that are able to sense those narrow tracks. You need the interfaces to be ... you need the head-disk interface to support these new components.

So, the distance between the head and the media has to be reduced. You need these coatings and layers on the heads and the media to be thin, to support that small spacing, but also to have thermal robustness and be able to protect the media under these extreme recording conditions. And the mechanics also have to support that very high track pitch.

So, for HAMR, yes, there’s the heat-assisted magnetic recording part, but the entire system has to support this brand-new recording mechanism.

Paul: It sounds like something out of science fiction, you know, when you’re talking about lasers heating up parts of that spinning disk. And can you talk to the level of precision, right? Because this is just so compelling when I learned about this.

Stephanie: If you take the distance between the head and the disk, you can’t even fit a DNA strand in between the two. You could fit the entire writer structure within a red blood cell. And, you know, these are analogies from probably 10 years ago. Now it’s all atomic-level precision.

All the components right now are on the order of tens of nanometers, so hundreds of micrometer-sized components. And yes, the manufacturing process has to support that aggressive scaling. The mechanical system has to be able to accurately track and place the head exactly where it’s supposed to be on the disk.

And tracks are only tens of nanometers in width. All of these systems continually work together to be able to achieve all of these high capacities from 30 terabytes, to 40 and beyond.

Paul: Yeah, I’ve heard. You and your colleagues when you’re talking about this technology is like, you know, atomic scale, engineering is literally that bit that you’re applying the heat to with the laser is less than an atom, an atoms diameter or something. The pinpoint key application is happening at that size, and it needs to heat it and cool it down. I think 800 Fahrenheit is the temperature that you’ve got to heat that spot on the disk and then it cools down again in a nanosecond or something.

Stephanie: Right.

Paul: The engineering is happening in terms of speed and size at completely different scale.

Stephanie: Right. Yeah. The heat spot is, yeah, 800 degrees Fahrenheit, only a few grains wide, a handful of grains wide. and that’s even now for Mozaic 3 and 4. But, well, once we talk about 10 terabyte per disk, those geometries get even more aggressive.

Paul: And it’s all, of this stuff as well. All of these nanoscales or, you know, these tiny systems and components. It’s not like you can go to a store or go somewhere and just buy them, right? They’re all bespoke, right? So, am I right in saying that we’ve custom built pretty much all of them, right? For this specific use case. Can, you, speak a little bit to that?

Stephanie: Seagate has committed to HAMR from the beginning, right? We’re the company that said HAMR’s the path forward. We had to design everything from scratch, right?Figuring outhe physics of the HAMR recording system, figuring out how to incorporate optical technology in write head. All of the media development in house to support the development of this new media. Everything is designed in house, so we have just a collection of amazing knowledge, right? That we had people that had to become optical, write head designers. And even now we’re exploring completely new designs that, you know, we’re just starting to develop and think about.

We may take information from what’s going on externally, right? We attend conferences, we fund research at universities, and we try to figure out what are the trends and what’s the learning going on in the outside world? We take all of those pieces and try to come up with, how do these new technologies ... how can we take them and make new heads, new readers, new media? And that’s all done in-house.

Paul: I mentioned this sounds like science fiction. It doesn’t seem real, but very much is. And it’s not an R&D project right now. Now we’re, you know, we are producing millions of these things.

Packaging them up as these little gray boxes and sending them out into the world. And what’s so interesting to me about the hard drive industry or what Seagate does is dealing with atoms, we’re dealing with nanometer scale engineering, but then it’s sotiny, right?

To be able to fit these, you know, more and more bits in this tiny little 3.5-inch box. But then we’re mass producing them, by the millions. I think, multi-exabytes of storage capacity every day, coming off our manufacturing lines. So, the interesting dichotomy is we’re truly engineering a tiny scale, but mass producing at an enormous scale. And now HAMR, we’re, doing that today. Can you maybe speak to that process like how can we produce that amount of exabyte output per day ... what goes into doing something like that? It is one thing to be able to produce one of these things. It’s another thing to be able to produce millions of them at the scale that we do and, store the world’s data.

Which is what we’re doing. So, it’s like everything from finding the materials. Building the components. There must be like millions. That we’re, producing all in house, like millions of these components, integrating them, testing them, like you said earlier, for performance and durability, all of the machinery, the manufacturing,

Stephanie: Right.

Paul: The people, the processes, all of theorchestration in doing that is just ...

Stephanie: Right.

Paul: A pretty mind-blowing operation to think about it like that.

Stephanie: With any of these technologies, you know, it started as an idea. For any of these new technologies integrated from generation to generation. It started as a research project, which may be one or two people. And then we add more people, the more promising the project becomes.

We start maybe using some of our existing tools to explore some of the concepts or maybe partnering with some external partners to look at whether or not the technology is feasible, and step by step, it moves down this funnel, where we have a lot of technologies on one end. And we evaluate all of them at a low level, and then the next step becomes, we involve more people in the process. We do more internal work. If the technology becomes viable of the few, that made it that far. Then only one or two go forward to the next stage. And then we start incorporating more of our existing processes ...

Into developing these new platforms. So, step by step, it moves from research to development, and within development, it just moves closer and closer to productization. So, it’s a gradual transition because you have to first understand the intricacies and the recording physics and all of these different things before you move on further down the maturity cycle. But then, we have a manufacturing process that is well established. And you have to modify that manufacturing process step by step to accommodate these new technologies. So, it’s certainly not a switch that you pull, right? And you go from PMR to HAMR?

It’s a very gradual process where it starts with a dream, an idea. and then little by little the company, more of the company, more of our people are involved. Until now it becomes this, thousands-of-people effort. And then eventually we can produce something to give to our customers.

Paul: So, you mentioned our customers. Why does any of this matter ... Seagate doing crazy stuff with quantum physics and innovating with these little boxes. Why does it matter to our customers?

Why does it matter to the world?

Stephanie: I think ultimately our customers are interested in higher capacity In a box that you can just plug in, and it behaves in much the same way and has good performance. I think they’re interested in the technologies that are going into this box because they want confidence that we have a plan for advancing the technology.

We’ve always projected that the world will create data at an exponential rate, and there will never be enough drives to store all of them. But we still have to keep increasing capacity because we do need to store a lot of the data that’s created. That’s becoming, more and more important.

So, ideas to keep increasing capacity within the same form factor are very important.

Paul: The typical kind of high-capacity stuff ... the Mozaic HAMR stuff. What are the kind of key environments that those drives tend to end up in and why?

Stephanie: All of our mass capacity drives, they end up in the big hyperscalers. They’re the ones that have limited footprint. So, we need to keep providing drives that have increased capacity and can support increasing the capacity available — without increasing the footprint of the data storage system.

We all know the names of these hyperscalers and our cloud service providers, and we all use their services, right? We all store a lot of data, and we generate a lot of data that we want to be able to access. So, it becomes important to supply a device that can support that: The growth in data over time.

Paul: It goes back to the idea of not messing with the form factor, because it’s got to slot into those. You can’t rip and replace or change up the format of these data center slots. You’ve got to stay … you’ve got to innovate within those parameters.

And yeah, like you said, the crazy growth in data — also the value of data —‚ the retention periods for that data accessibility. For data, all of these kinds of dynamics are driving up storage demand. And, yeah, I guess why it matters is just trying to help our customers keep pace with that growth curve.

An interesting kind of customer scenario that I was looking at yesterday is just the idea of when you’re at this kind of fleet scale. Where you’re operating a fleet of drives into the hundreds of thousands and beyond: the impact. When you’re at exabyte scale, an upgrade scenario from maybe a fleet of 20 terabytes drives to a fleet of 40 terabytes is you’re essentially doubling your storage — your raw storage capacity — within the same physical footprint. And you alluded to earlier, one of the challenges today is just space physics, right? There are limitations to physics. Just like there are limitations inside the drive, there are limitations on the environment that these things slot into. And so why, is aerial density important?

I think that kind of that speaks to that.

Stephanie: Everything that we’re working on is just pure aerial density. I mean, not all of it, but a lot of what we’re working on is how do we keep increasing aerial density by just decreasing or increasing capacity ... by decreasing the size of the bit. So, all of these technologies can really fit within the box size that we have. I think the way forward is being able to keep increasing capacity to a hundred terabytes. Because it’s pure aerial density.

Paul: You mentioned something earlier that caught my attention. Ten terabytes per disk, right? So Mozaic, we’re at four, four plus, right? And you’re looking future state, how do we keep innovating with aerial density to increase capacity over time? What are some of the ways that you guys are thinking about that, you can share with me?

Stephanie: Absolutely. Yeah. we don’t see any fundamental roadblocks from between four beyond right to 10 and beyond that. Like I said, it’s all about increasing pure aerial density. So that requires decreasing the size of the components. So, the reader has to be smaller.

The critical write elements also have to be smaller, and the grain size has to be smaller to support smaller and smaller bits. So, we’ve actually demonstrated in the lab seven terabyte per disk. So that’s double-ish from where we are in product today,

Paul: Cool.

Stephanie: And this is in a real lab, so we’re actually taking next generation heads in media and recording real information on the disk and recovering that information. The components ... they’re much more aggressive than what we have today in terms of geometry. So that system, that seven terabyte produced system is not immediately productizable, but it serves as a proof of concept for what is achievable in a recording system. Another thing that’s in the demo is multisensor magnetic recording.

So, we do signal processing to mimic having two readers, and that allows us to use narrower readers. But we need two of them in order to resolve the information coming from narrower tracks. We’ve proven seven terabytes per disk.

Paul: If you have one reader today ...

Stephanie: We have one reader today in HAMR. Yes. So, if we put two of them, we could use narrower readers than we would otherwise be able to.

Because like I said, magnetic materials become unstable at small volumes. so yes, with HAMR we can write really narrow tracks, but the reader also has to be narrow. If we want to get to higher and higher aerial densities, our readers also have to be narrow. Using two or more readers, we can actually reduce the size of the reader than we would otherwise be able to. So in these aerial density demos, there’s also mimicking multisensor magnetic recording. So again, we can use narrower readers than we would be able to put in a product today. but these demos, you know, demonstrate we could write bits, that are small enough to support seven terabyte per disk.

Now it’s a matter of actually producing these narrow features, narrow components, in a reliable way. Now, beyond seven terabytes produce, there are many more ideas. So, taking different multisensor magnetic recording technologies, we can do two-dimensional magnetic recording. So that’s a different way of encoding the information on the disk.

I still need two readers. there’s this other concept called vector recording, where I’m sensing different field directions that are coming from the written patterns on the disk. And ideas like these are able to relax the reader width, scaling, issue. I can use wider readers than I would otherwise be able to.

We also need new mechanical systems to support this very high track pitch and be able to support, these multisensor magnetic recording technologies that require very precise, distances between the two readers. And we’re looking at new reader materials, new reader designs. We continuing to scale the critical dimensions of the HAMR writer, continuing to reduce the media grain size, use new media materials.

To support smaller grain size. so yeah, lots of ideas to get the 10 terabyte per disk. Still a lot that needs to be worked on. we’re partner, like I mentioned, we partner with a lot of different universities to explore different concepts. There are new ideas that we may not have the expertise to explore in house, that we, partner with different researchers and, they can look at these, technologies and we might eventually incorporate them into our designs. So, there’s still lots that needs to be figured out to get to 10 terabyte per disk and beyond. But we think HAMR is highly scalable.

Like the HAMR is a very, very good framework to build new technologies off of. It just fundamentally supports, capacity and aerial density growth. And then it’s just a matter of seeing how far that will take us, which is still an open question, but we’re confident that we can certainly get to 10.

And beyond that, we’ll have to see how far our HAMR can take us.

Paul: The HAMR paradigm holds like beyond like to 10 terabytes per disk, and we think beyond. It’s just, it’s again, the principle of just shrinking, the components and the systems and innovating around that nano scale engineering in order to get to those aerial density milestones.

Stephanie: Yeah. Yeah, always. reducing the size of the components is, key, right? that’s what derives aerial density. and you know, also thinking of new ways of recording the information on the disk. so how can I record more information while relaxing the bit size, while relaxing the different component geometries?

Those are also things that we’re thinking about. But yeah, I think ultimately HAMR, is able to achieve, yeah, 10 terabyte per disk and beyond that and, you know, it’s, just amazing that, you know, hard drives have been around for many, years and you know, you have one there, and from the outside the box looks the same today as it does, you know, for that box you have there.

But if you open it up, it actually also looks the same. Right? When people first designed a hard drive, did they really think that same design, would be able to support 30, 40, a hundred terabytes? Because yeah, not all you need to do, but you know, it’s such a well-designed system that it supports this extreme geometrical scaling. The mechanical systems, all of that can support these extreme geometries. Our goal is to maintain that rotating disk architecture for as long as possible.

Paul: Do you have like beyond? Beyond that, that, you know, age old, you know, rotating disk architecture that has, you know, survived multiple technology transitions and is still storing the world’s data today. Do you have research projects or ideation outside of that kind of core innovation path? Are there other interesting recording technologies that are interesting to the company?

Stephanie: Yeah, our fundamental goal is to keep extending that rotating disk architecture and even beyond magnetic recording. So eventually we’ll hit that superparamagnetic limit and we won’t be able to keep reducing the grain size. When that would be, we don’t know. But there are other types of materials that could replace magnetic materials.

There’s ferroelectrics. There may be others that could serve as a recording medium that could store information at smaller bit sizes than magnetic materials are able to. So that’s all very much a very speculative research project. And it’s in the realm of just doing fundamental physics modeling — partnering with universities. Beyond rotating disk architecture, we like to look at things like DNA data storage. We don’t, aren’t just looking at trends in other non-HDD type architectures. However, there’s really nothing that we are looking at that will be an HDD replacement. HDD sits in a very specific place in the data storage hierarchy.

And, you know, none of the technologies that are being talked about are HDD replacements. We’re trying to see what can go into the HDD to continue increasing the capacity, of the device.

Paul: Yeah, it is almost like these are adjacent fields of research that we would look to try and integrate into our core architecture. Is that, right? Is that why we study it? Just to see are there things that we can learn and pull into our core roadmap?

Stephanie: Well, I think Seagate should be involved in more parts of the datasphere than just HDD, right? We need to figure out if there are other opportunities for us to be involved in these other parts of the datasphere.

Paul: I think we need to wrap up. Stephanie, it’s been such a fascinating conversation. What didn’t I ask you that’s important for our audience to know about this technology and the work that you do?

Stephanie: Why would I choose to work in magnetic recording? Right. because that’s not a typical field that, you know, grad students back in 2010 would go into. But as a student, and even as I became an engineer at Seagate and I was learning about all the technologies, it’s amazing how many diverse technologies are in a hard drive, and how much these technologies have evolved over time. And I don’t think that’s something that’s clear to the world. Open box, it looks the same on the outside, and theinside, but if you take a microscope or something stronger than that and look at all of the different components, they are completely different materials.

Completely different physics used to do all the things that a hard drive does. and I think, people should know that, you know, a hard drive is much more than what it looks. It’s amazing technology. yeah, that, you know, it’s just amazing how much it scaled over time or changed over time.

Paul: Yeah. It’s like the allure of, you know, working on a startup that does nano robotics, for example. It’s like that, sounds, you know, like something, an engineer, like an up and coming new, you know, fresh grad out of school would be super interested and compelled to work on. That’s a hard drive, right? It’s a robot. It’s a nano robot that is, delivering incredible value to the world. And yeah, like you said, like it’s such a well-designed system that what you can see with the naked eye actually hasn’t changed in a few decades, but it’s like, it’s what’s on the inside, right?

It’s under the microscope. Then you’re, so many incredible innovations and technologies being brought together and integrated and then manufactured at Mind blowing scale. Really, phenomenal.

Stephanie: Absolutely.

Paul: Stephanie, it’s been such a pleasure chatting. I learned a ton about the work that you do and, you really helped educate me on some parts of what the company does that I wasn’t quite aware of. So, I appreciate you walking me through that and appreciate your time today.

Stephanie: Thank you so much for the opportunity. It’s been a pleasure chatting with you, too.

Paul: Yeah, can’t, wait to see those next innovation nodes and hitting seven terabytes a disk, and 10, and all of the work you and your team and your, colleagues are driving here at Seagate. Super impressive.

Stephanie: Thank you so much, Paul.