Best's Biography

JOHN BEST

First, a brief look at the early history.  [Video Clip]
 

Since I am going to talk mostly about hard drive technology I wanted to show a little bit of this, its some of the same images Al had, but you can sort of get a different picture of it when you see the motion itself going on. [Video Clip] The most important single parameter advancing in this whole industry has been areal density because that’s the thing, as Jim was just saying, that has advanced over many orders of magnitude and, in fact, from the days of the RAMAC to shipping products today areal density of hard drives has advanced by about a factor of a million.  So, the real question is, as we go forward, are we going to keep advancing and how much farther can we go?      From about 1956 to 1991 the rate of advance was about 30% per year.  Since 1991 the average has been about 60% per year.  All exponential changes have the characteristic that eventually they die or they consume the entire universe, so we expect sometime in the near future, or at least in the next forty years this advance is going to end.  So we’re going to talk a little bit about what can happen.  So will there be demise of the hard drive?  If we’re going to think about the end of hard drive technology what we might do is think about what’s happened with the other magnetic recording technologies and why they’ve become extinct.

 

We can take cases of the wire recorder [Video Clip] or steel tape recording that may become extinct because they didn’t scale very well. In the case of steel tape, as we heard earlier today, they also could be dangerous and they’re inherently expensive, which is the case of the magnetic drum, which also didn’t scale terribly well in terms of volume density.  Core memory is a different case, though it’s not really strictly magnetic recording technology, It’s magnetic technology.  It was faced with competition from a competing technology, that being semi-conductor DRAM.  Some still question magneto-optic disk as to whether it will keep living on forever or will slowly die away or continue to grow, but it again is threatened by the very magnetic recording technology of hard drives and by other optical recording technologies.

 

Let’s look at enduring magnetic recording technologies and why they have continued to advance.  Tape technology: audio and videotape.  One is, that the media itself is incredibly cheap.  You can get very large areas packed in a very dense space very inexpensively.  Second, if you look at extending tape technology, digital recording technology has in fact been very scalable, and therefore can move forward and continue to advance.  Diskettes, again is a case of being very inexpensive and driven by standards of interchange, so therefore have lived on for a very long time.  Some question arise about how that will advance and whether it will continue into the future.  That leads us to hard drives.  Hard drives have been very scalable in terms of continuing to advance over that forty-year period.  A factor of a million advance to something is unprecedented in the history technology with the possible exception of the semi-conductor industry, in terms of many orders of advancement over time.

So one of the questions is, if we keep advancing will we be able to in fact use all of the storage that hard drives can store if we go way out beyond the 2001. The amount of capacity for all the disk drives made by this industry starts to look pretty astounding.  Today the total amount of data sitting on-line in hard drives is about 300 petabytes (petabytes being a million gigabytes).  Off-line, not on hard drives today is another 8 exabytes of data that is currently digital, which if the cost keeps coming down, there is no reason not to put this data onto hard drives, being much more convenient, easy to access and everything else.  So it’s data that’s already digital that’s there ready to move onto hard drives.  But that will consume only about another factor of 10 to 20 of storage.  Then there’s another set of data that’s sitting on other kinds of media that to date not digitized and that’s currently analogued.  This is the situation of much existing data today.  And there is another 200 exabytes.  So there’s the factor of a thousand between the total amount of storage on hard drives today to data that’s sitting there out in the world that could easily be digitized and could be very productively used on hard drives in an on-line manner.  So there’s no lack of demand and this doesn’t include the fact that if the cost goes down low enough, and we’ll talk about this later, what would happen if you start to digitize things that aren’t even digitized today such as the total sum of people’s experiences, everything they see and hear, you could imagine this information being captured onto hard drives as it gets a low enough cost. One would never have the excuse of forgetting things in legal depositions.

Other kinds of things that go on and again open up new market opportunities, or new form factors, that enable completely different things to happen.  We talked a little bit about the idea of a micro drive.  This is just a little video clip that shows some of the applications with emergence of semi-conductor technology that enables you to make high resolution digital cameras, both video and still.  [Video Clip] These very compact kinds of drives which are just now beginning to have enough capacity to be interesting in those applications.  It starts to be a very interesting market, which again will capture much more data in a digital format then was done previously.  So again we’ll see more and more capacity, in initial drives there will be 340 megabytes and continue to advance from there.

So that brings us to the main topic here, that is the areal density versus time and will it continue to advance at the 60% rate, as you can see here it’s been doing that since 1991.  A couple interesting things here, there’s a set of points, the blue points there that are demonstration values that we did in the research lab at IBM before moving them in to product.  At 1, 3, and 5 gigabits per square inch you’ll see we were at a pace there that didn’t match the actual advance that is going on in products.  So, in the 1997 time frame we got worried that the products were going to ship before we demonstrated the areal density.  The beginning of last year was when we sort of moved up the pace of developing the technology itself and demonstrated 10 gigabits per square inch at the beginning of this year.

Now there are a number of factors that go into enabling that advancement.  The real thing that has been used throughout the history of the hard drive is some very simple scaling laws, and that is the magnetic system itself. If you simply shrink absolutely everything, that is the thickness of all the media, all tolerances, all the dimensions, all the mechanics, everything the drive basically still works.  The magnetic recording process still works, the transitions get sharper proportional with shrinking everything else and it still works.  Of course this in not an easy thing to do.  It requires vastly improved processes, lots of inventions require higher precision.  The one thing that does happen is the fundamental signal to noise drops.  It doesn’t scale as the drive shrinks so it requires new sensors and materials.

Now the real beauty of this technology is that it is not a single technology no matter what you’re interested in you can be involved in magnetic recording and hard drive technology.  And you have to advance all of them to keep this technology moving forward.  From mechanics in the kinds of things that position the head over a disk, the head sensitivity as we talked about that’s critical to maintaining the signal to noise, electronics both in being able to read and write very quickly because as we store data more densely if you want to get to it quickly you have more data flowing under the head at any given time so the data rate also advances rapidly as you scale the technology forward.  There’s all kinds of computer science and everything else that goes into setting the channel data rate.  Head disk spacing, we need to scale that space between the head the fundamental media itself where the data is stored.  Now we’re going to talk a little bit about just a few of these to see what are going to be the limits to continuing to advance the pace of areal density.

First starting with the mechanical components.  Again, as we’ve advanced the areal density 60% per year that means that the track density needs to advance about 30% per year if we shrink everything proportionally.  In fact that’s been happening at the same time we’ve been speeding up the whole process, rotation rates have been increasing so the rotational access time decreases, speeding up the access time of the drive. Partly as we shrink the mechanics all mechanical resonances go to a higher frequency, this enables you to servo more accurately.  The mechanical scaling that we’ve done, going from 14 inch disk drives all the way down to 2.5 inch disk drives, has been to our advantage in both speeding up the mechanical processes as well as making them more precise, thus positioning more accurately.  It’s part of the formula for enabling the areal density to increase.  There is nothing fundamental here to keep this from moving for quite some time to come.

In fact the set of things we’ve been doing, we’ve talked about one, first, shrinking the mechanical components.  As we continue to advance, we learn how to optimize much more effectively.  How we designed, controlling the modes of vibrations, and eventually going to more exotic technologies such as using silicon micro-mechanics as secondary actuators.  Here is a prototype actuator made of silicon that sits between the suspension and the head so with very fine control you can position the head accurately under a track.  This kind of thing can extend us for easily for several orders of magnitude more track precision than we have today.  This isn’t the limit that’s going to keep us from scaling looking forward.

Second is magnetic and physical spacing.  This one’s interesting if you look at the history of people predicting the demise of hard drive technology.  Throughout the 70’s and early 80’s there were a number of people who predicted that we would stop advancing the areal density of magnetic recording and another technology would take it over. Almost always the predictions have been based on the fact that you can never foresee the head getting closer to a disk than would be required for moving the technology out about three or four years from where it was at that time.  It’s just a tough engineering problem.  You need to control thinner overcoats, air bearing design, etc. But it’s not a fundamental issue until you get out to the kinds of spacings that are an order of an atomic diameter, which is a tenth of a nanometer.  That’s in fact a long way out and not a constraint in the immediate future from a fundamental standpoint of how far we can advance this technology.

We can look with a little more detail at some of the things that we do to continue to advance the disk and decrease the spacing.  One is we decrease the surface roughness of the disk.  That has the advantage that the disk noise goes down and enables us to fly closer to the disk without wear and contact.  In the process the durability of the interface improves, the magnetic properties of the disk improves, but the big problem is stiction between the head and the disk, when the head comes to rest on the disk. This is a constraint on how smooth the surface can be. Therefore, you might think it a constraint on how far we can extend the technology.  Several things have been happening; one is that we have moved to smaller sized SLIDErs so we can somewhat reduce stiction.  But also, as we look at how we reduce stiction we’ve gone to a number of different approaches.  One of them is more sophisticated techniques for inducing microroughness in the disk.  Things such as sputter texture to give very well controlled fine texture, laser texture which can be applied at a landing zone only, so you can have very smooth disk over the main data surface but land on an area where you have controlled roughness to prevent stiction. [Video Clip]  [Video Clip]

And then finally, you can go to a load/unload mechanism in which case the head never comes to rest on the disk so it really eliminates the stiction problem. [Video Clip] So in the case of scaling to low flying height, one piece of it being the disk roughness, by invention and continuous engineering improvement, we’ve continued to make breakthroughs that have advanced this technology forward.  Again this factor doesn’t appear to be over the immediate future of the next five to ten years the real limiting constraint on how far we can move forward the technology.

So the next thing is the heads.  As I’ve said before one of the keys to the scaling technology is you need more sensitive transducers in order to be able to maintain the signal to noise.  We go back to the early heads; they were both cores of magnetic ferrite with a real wire wrapped around it.  That had a limitation mostly in fabrication and scaling to smaller dimensions that the introduction of the thin film head addressed. We applied with photolithography processes that enabled you to scale and enabled you to open up new materials to use and made it possible to have more complicated head structures. Separate read and write structures enabled the introduction of the MR head in 1991.  The MR head was fundamentally a more sensitive transducer than the inductive read head and that was the key to enabling the kind of scaling that’s gone forward from 1991. That was really one of the key pieces to increasing the rate of progress from 30% per year to 60% per year in areal density.  Subsequent to that, beginning at the end of last year we’ve introduced the giant magneto resistance head which looks physically, from a manufacturing standpoint, very much like the MR head, but it’s a whole new physical phenomenon and is again much more sensitive. The GMR head has a long way to go in terms of advances in sensitivity. And there are other magneto resistance type effects that have even larger total sensitivity than GMR that are on the horizon that might eventually be put into read heads.  So again, there has been a tremendous advance in technology over the last 40 years in heads.  There are tough problems to solve in continuing to advance heads, but nothing fundamental that would make this the limiting factor on advancing this technology further.

So that brings us to the thing that we all believe is eventually the basic physics that you can’t get around and that it really is fundamentally thermodynamics.  If we look at an idealized track here, at a density comparable with today’s product (the media today no longer of course is iron-oxide particles in an epoxy binder, but in fact is a thin film sputtered cobalt transition metal alloy).  The media itself and the bits really look like a collection of small somewhat independently and mostly decoupled magnetic grains.  As we shrink the bit, in order to maintain a constant signal to noise ratio, we need to shrink those magnetic grains so that there’s the same number of these randomly oriented grains in a given bit as we shrink the bit.  So if we think about that, each one of those individual grains has an energy barrier to being spontaneously, or under an applied field, reversed to the opposite direction.  When we apply a large field to reverse the between two directions of magnetization, that’s how we write the data in the first place.  Once we write the transitions (a transition, or bit, is a location where the magnetization is switched) the internal magnetic field is in a direction to try and reverse the resulting magnetization, or demagnetize the bit.  The problem is as we make the bit smaller and smaller and reduce the medium grain size, the size of this energy barrier to the reversal of this magnetized bit of a bit reversing gets smaller and smaller. In fact it gets smaller and smaller very quickly so that what can happen is under thermal agitation at room temperature is that the magnetization will spontaneously reverse and you lose all the information.  This phenomena is a very steep function of the volume of these bits, in fact, the time scale over which that reversal occurs is exponential in the volume of the bits.  And with today’s materials and the anisotropy energy associated with those materials, we find that at a density at the order of 20-40 gigabits per square inch that this is going to be a severely limiting factor. It’s hard to play a little trick to advance a long way because if you think about this exponential decay,  if you shrink the grain size from one that’s very very stable, that is has enough energy associated with a strongly coupled grain, to one that’s a factor of 2 smaller and you can cross a threshold of the exponential, where the time can change from hundreds of years of stability to nanoseconds in which the spontaneous decay occurs. This is already being seen in today’s materials.

So what can we do about this?  First is to continue the scaling as we’re doing it today with the basic material set we have today like cobalt transition metal alloys, and that again will get us to the order of 40 gigabits per square inch. Beyond that we change the aspect ratio of our bits.  Today the track width is about 16 times wider than the length of the bit.  If we go to a much more square bit, say 4:1, then you can trade off slightly thicker media, or larger volume of the particle for the same number of particles in the bit. The effect is that you can gain about a factor of one and a half in areal density before you get to the super-paramagnetic limit problem.  It has some challenges with it because now things that go on at the edges of the track which affect the signal to noise and the mechanical tracking has to advance more quickly than it has in the past in order to go down this path of improving areal density.  As we’ve talked about before, there are ideas and prototypes today for ways that can enable us rapidly advance mechanical tracking.      Second, is to just go to different materials formulations that are ideas that people ins various laboratories are looking at for higher energy media might be able to get another factor of 2 that way. The challenge is that write heads can’t produce enough field to write the media so we’re probably going to need to change the materials in the head as well.  Challenging, difficult, may take some time, but probably is doable.  And finally we need to learn to actually operate with lower signal to noise.  We can do things like run longer block sizes with much more powerful ECC’s so that we can maintain a user error rate that is very, very low with lower signal to noise.  So all these things are difficult challenges, but basically straight forward from a fundamental physics or engineering standpoint and can get us into the 100 to 200 gigabits per square inch regime.  Which is about as far as anybody has demonstrated with any other exotic recording scheme that they pretend might replace magnetic recording.

Now beyond that we can think of more radical things to do.  One is keepered media. Put soft magnetic material on top of and on the bottom of the storage layer in order to reduce those internal fields that try and reverse their magnetization.  This is tricky because it prevents any external field so there’s no field for the head to read so you have to, as you’re reading, apply a field from the head to saturate the top keeper layer. Whether that really gives an advantage or not no one knows.  Second, we can go to perpendicular recording where again the internal magnetization stabilizes a bit. This may give another factor of one and a half or two or but it still suffers from the same fundamental principles of super-paramagnitism.      Another idea that has come along is thermally assisted recording, that is, using a media that has a very, very high energy at room temperature, so it cannot be reversed at all.  It’s very, very stable and as you go to write you just heat it up, which reduces its coercivity, and apply the reversal field then move on.  The only problem with this is it suffers from the problem of the exponential dependence of the time factor on the bit volume.  If you decrease the volume very much then, even in the time during which you heat this material and then cool it again after you reverse its state, it will self demagnetize again in nanoseconds.  This approach is probably not good for another factor of more than 50% or a factor of 2 either.  These things might help but not be clearly advantageous.

But there’s one thing we can do that from a fundamental principle standpoint could have a huge advantage.  That is, change our individual bits from being made of many small independently magnetized particles to a single strongly exchange coupled particle. Here the volume of the entire bit becomes the thing that has to be large enough so that it’s magnetic state is not spontaneously reversed.  This would be referred to as the discreet bit recording or patterned media.  Here there are lots of tricks to make this work.  You have to actually figure out how to pattern the media.  You have to figure out how to orient the anisotropy of the media along the track that you’re writing.  You need to synchronize the reading and writing with this pattern.  There are a lot of things that need to be done, tough engineering problems, tough materials problems to make this work, but fundamentally from the media stability and thermal noise standpoint it can get you to the 10 terabits per square inch regime.  Realistically, other factors such as the spacing between the head and the disk and other things will probably limit it to something along the order of a terabit per square inch.

So terabit per square inch compared with where we are today, that’s another factor of about 200-500 in areal density from where products are today.  Given all the alternatives that we’ve looked at, such as using atomic force microscopy to deform and make pits and then read them back in a plastic substrate, it is tough to get more than about 200 gigabits per square inch by that method, while I think that is fairly straightforward for magnetic recording. Holographic storage, the idea’s been around for a very, very long time.  It’s always been looking for the ideal material.  It’s never seems to come and even if it does it’s not clear the density will be that much greater than magnetic recording when you look at stacking things, the cost of the head and everything.  Near field optical recording is another one that again seems to be limited due to the optics to again to a 100 gigabits per square inch kind of regime so it’s facing competition with hard drive.  So I think that probably isn’t going to displace the hard drive.  Finally the one thing that eventually will is writing with individual atoms one by one, which can get out another million times in areal density.

That gives you the scanning tunneling microscope where you can not only image individual atoms but actually build structures. [Video Clip] The top structure in the picture is atoms of iron individually placed on a copper crystal and imaging the electronic wave form of the electrons on the surface of the copper in the middle.  The one at the bottom is electronic graffiti with xenon atoms on a silicon surface.  So, again with the exception of the atomic storage, which we have no idea how to make practical and is a long way off, there’s really nothing to displace magnetic recording for on-line high-performance storage with a hard drive.  Today we are demonstrating about 12 gigabits per square inch.  We see straightforward ways to get to 20-40 gigabits per square inch.  Ways that aren’t quite so straight forward, but look pretty practical that get you to the 100-200 gigabits per square inch, whether we can keep up the 60% compound growth rate in that regime isn’t so clear.  It might be a little slower because it gets harder.  And then we’ve got some ideas that’ll get us out to a terabit per square inch and again the pace at which we can do that I think is unknown at this point, maybe it fall back to the traditional 30% per year, but eventually it will happen.  Then finally out past that we will see something like atomic storage.    [Video Clip]