Although the heart of HDD technology is made of simple principles, it required decades of development and scientific researches, enormous number of complex, non-trivial, sometimes ingenious and incredible solutions for it to reach such heights. All that information is little-known outside the circle of people involved in hard drive manufacturing processes. We have discussed it with the right person for this task and were able to ask all possible questions related to hard drive technologies that are being used and will be introduced in future. Please welcome: Alex Blackwell, chief engineer of Western Digital in EMEA region.
Mr. Blackwell often interacts with computer mass media but its not the case when a person has to do it only because it is a part of his job duties. You can feel it because he really enjoys telling people about technologies. Alex is so involved into the conversation that two hours of our interview passed in one breath. And I cannot call it an interview as such. I did not have to get out information from him I asked him several questions and got a very broad and most interesting information than we expected to get. Eventually we’ve got a complete lecture on interesting and unobvious facts concerning hard drives.
When making the list of question we tried to avoid such commonplace questions like “how is WD these days and what are its plans for future?”, so we tried to find out as much information as possible concerning hard drives at large without appearing naive and ignorant. Alex with pleasure allows his conversation partner to be as eager to learn as possible.
On top of the above Mr. Blackwell’s speech is bright and full of metaphors and humour. We will try to convey it in the text by making it as close to ‘undressed’ report verbatim as possible. Nevertheless, since the conversation revolved around same questions we will compose it in a way of a syllabus comprising of several main topics. No single plot, just a collection of stunning stories about hard drives. Everything written here is the direct speech of Alex Blackwell, all questions and comments are in italics.
On head landing and integrated generator
Q: We have recently found out that hard drive uses electrical generator to complete writing to sector in case of emergency. Can you tell about it in details?
A.B.: When there is a short circuit, first and foremost thing for the safety of actuator is to land heads. Because if they land onto magnetic carrier they would simply stick and won’t be able to lift any more (when in operation the head in fact flies over the surface due to air flow). That is the end. Such smooth surfaces it has. Imagine two absolutely smooth appressed glass sheets. How much force is required to break them apart! If you turn off the drive after heads stuck to disk, then rotating spindle would simply tear off the actuator’s tip. That is why in order to land we lift heads and park them onto a separate plastic area. Or more precisely, we land actuator and heads hang off the tip in the air.
Actuator’s tip ‘dropped’ on platter
When short circuit occurs we always have some time to land the heads. This operation is carried out with the help of electrical generator. But there is no generator as a separate device in a hard drive. The motor is simply used in ‘reverse’; this can be done with any electrical motor.
This is the state of things of the last 15-20 years. Older drives parked their heads directly onto the disk surface, on its inside edge. There was a magnetic lock that held an actuator on site. If you remember, when you turned such old drive, you would hear a click, meaning that actuator approached a magnet and latched there. Western Digital stopped manufacturing such drives in 2005–2006, perhaps even in 2007.
It was possible to park heads directly on the disk because originally the surface was not that smooth and heads were bigger. Frankly speaking, back then everything was easier. Later on we had to make surface very smooth to enable head flying very close to it (today, distance between the surface and head is several nanometres). And once it became too smooth to fly from it after parking. Then we started using laser to create a texture on disk surface in landing zone. Now, since 2007, landing zone is outside the disk surface on a plastic area. In other words, head landing principle has undergone three development stages, but, despite this, a lot of fine technologies are applied in this sphere.
But let’s get back to short circuit situation. In addition to parking heads we have a second task – saving as much user data as possible. We need to transfer to carrier a piece of information, written at the moment, and finish writing to current sector. For that we simply use residual rotation of carrier.
Some impressive numbers and dual stage actuator
The first hard drive appeared in 1956. Remember other technologies from 50-ies. For example, radio valve. Since then we had transistors, then integrated circuits and then Large Scale Integration (Large Scale Integration are microcircuits containing hundreds of thousands transistors). Or take audio recording. We used platters with rotation speed of 78 RPM for most time. At first with plastic needles then with diamond and then magnetic tape, CD, MP3. Some of the technologies simply jumped ahead but disk drives still work as they worked before. There is a spinning disk and actuator moving along it, magnetic surface with inductive read/write principle. So, most probably cars are left as they were before.
But imagine the first hard drive made by IBM. Let us assume that the size of one bit on this disk of 50-ies is comparable to the New York Stadium. So how big is a bit on modern drive? Is it a size of this table? Is it a size of this room? Or is it a size of a thumb? Correct, a thumb! Areas occupied by one bit now and back then are in scales 108. That means 104 in each direction.
IBM 350 (1956) — the first hard drive. Designed for IBM 305 RAMAC (courtesy of Wikimedia Commons)
A geometry of hard drive constantly shrinks. Now tracks on carrier are located at a distance of 50-60 nm from each other. And now remember Intel microprocessors that use photolithography, factories with gigantic equipment to manufacture 28 nm norm. At the same time we have a spinning disk and we can position head in the centre of one of the tracks that are separated by 60 nm only within the accuracy of 10 nm. It is a real hi-tech.
Do you know what a dual stage actuator is? Imagine that my right hand is an actuator with a head at the end. Here is the turning point in the shoulder joint. So if you need to improve hand positioning take notice of the finger joint. Dual stage actuator has a kind of an additional little actuator that can move several tracks left and right. Due to this we can increase positioning accuracy. We have been using this technology for around several years in corporate products (RE3 series) and in 2012 we introduced it in some consumer models. In Green series terabyte drive, in several Blue, in all Red line and now in Black as well.
Design of dual stage actuator (United States Patent 6624983)
WD Black and terabyte platters
Q: Please tell why WD Black series drives show such impressive performance, in random access tests in particular?
A.B.: One of the bases of high performance is a spindle speed. Second basis is fast actuator due to which track seek time is reduced. In WD Black and RE drive series actuator motor contains two big magnets. Stronger magnet allows moving heads faster. In other series, Blue and Green, only one more compact magnet is installed, that is why Black is faster than Blue in terms of random access speed, though the latter also operate at 7200 RPM.
Q: When shall we expect WD Black drives with 1 TB platters?
A.B.: It depends on priorities. There are no obstacles in terms of technology preventing us from doing it. Terabyte platters are already used in Green series at 1-3 TB capacities and in Blue series. You know, when you design a hard drive and want to sell and get profit, then you need to combine a number of parameters: performance, capacity, component yield ratio during manufacturing process and many more. A combination of factors is important, not only possession of a certain technology. I suppose that terabyte platters have not yet entered the zone of effective features for WD Black.
WD Black (left) and WD Blue (right) — check the magnet size difference
Q: What are GPP / GMR heads (Perpendicular to Plane / Giant Magnetoresistance), that are used in today’s hard drives? How do they work?
A.B.: Original IBM hard drive and other drives until 1996-1997 had single read/write heads. Such head is a disrupted ring with wire wrapped on top. When a current is supplied to the wire it generates a magnetic field that ‘drains’ through the gap in the ring. If you near this gap to something that could be magnetized, it magnetizes. This is what happens to platter surface in hard drive: appearing areas with magnetic poles – North and South. On the other hand, if you do not apply voltage to head and instead simply move it along magnetized area, it generates current in the head.
Actuator and its tip under the microscope (photo courtesy of Andrew Hazelden, www.andrewhazelden.com)
It became clear with time that a single device represents a compromise. At that time the idea of magnetoresistivity was introduced. They started using a resistor that changes resistance in presence of magnetic field as a reading head. And as a writing head they started using a separate inductive part. And no more compromise hereafter. Later on the second generation of this technology appeared — GMR (Giant Magnetoresistance), where Giant points to the voltage value that allows developing resistive element. It simply became more sensitive. And for the future after GMR we have this thing: TuMR — Tunneling Magnetoresistance that would increase head efficiency even more.
Now about recording. A coil with gap in the middle, that I talked about initially, is used for so-called longitudinal magnetic recording. Magnetized areas on a platter are formed lengthwise. It is like cars are being parked on a street.
Longitudinal and perpendicular recording
But now we take and install these magnets vertically. We get perpendicular recording. Without knowing the technology it is difficult to imagine how it actually works. In fact, you need to add one more layer to magnetic platter that is sort of reflecting one of the coil poles and creates a weak magnetic effect spread over a large area. This is how perpendicular recording works. It would be better for cars to park vertically as well, especially is Moscow. Main thing is not to forget removing coffee from cup holder.
Future technologies: Shingled Recording, Heat-Assisted Recording, Bit-Patterned Media
Q: From time to time we see in the news such technologies as Shingled Recording, Heat-Assisted Recording, Bit-Patterned Media that will allow increasing record density on HDD in future. What are they and how soon can we expect them to be available commercially?
A.B.: Shingled Recording is a kind of a strange technology that borrows something from SSD. The idea is to record data to magnetic platter consecutively; only instead of one track here and one there with a distance between them we will lay tracks one over another. But we will have to be very careful in how we apply this technology because Shingled Recording will influence the way the host controller will be using the drive. Eventually it can turn out to be only a niche solution for a specific market that will be interested in it.
Auth.: Since track are laid one over another, when you need to record data fragment to one track, you will need first to read and save all data from crossing tracks and then write it all back to disk. Second option – try recording to free areas that will lead to file fragmentation. All that really reminds SSD with NAND memory that also allows freeing cells for rewrites only as large blocks (say, 128 Kbytes each). Since hard drives are not good with random access, it looks like that destiny of Shingled Recording lies in storing huge data volumes predominantly with consecutive recording.
Another technology is a Heat-Assisted Recording. Lets take a piece of butter. You take it out from the fridge and put it on a table on a warm day. Later it becomes very soft and you can make a hole in it with a finger. But to make a hole in a piece from the fridge you need something really sharp. This way is good for imagining the property of magnetic carrier called coercivity.
Coercive force — a measure of the magnetization of a ferromagnetic material as expressed by the external magnetic field strength necessary to demagnetize it.
Soft butter suits to low density drives (low coercivity) and thumb (large distance between head and carrier). Hard butter is required for high density drives to hold magnetized areas (high coercivity) and to record it needs a very, very sharp finger: high magnetic field intensity and smaller gap between writing head and carrier.
Let’s take IBM hard drive from 1950s. We could see wires wrapped on the head and distance to carrier. Now heads are made as integrated circuits made of thin-film materials. But I will have to kill you if I tell too much about heads. Heads is the technology that keeps us afloat in business.
So, we can develop only limited magnetic field intensity and will not be able to near the head to he carrier. But we can make a magnetized carrier more ‘solid’ We need a small laser to make the butter soft, make a hole in it and put back into the fridge. You can expect this technology to be commercially available within two-three years from now. But first – Shingled Recording that can be combined with Heat-Assisted Recording as well as applied independently.
Heat-Assisted Magnetic Recording (HAMR)
Another technology in our technological roadmap after HAMR is a Bit-Patterned Media. HAMR still implies using continuous carrier, though, made of exotic materials. Bit-Patterned Media means recording of individual data bits surrounded by void. At this, magnetized area can be situated very close to each other without interfering. The problem is in creating structures of this size. Existing lithography does not give us record density that we have with a continuous carrier. Huge changes should take place in lithography to increase record density with the help of Bit-Patterned Media.
Native support for 4K sectors
Q: So when then?
A.B.: Would you like to see native support for 4K sectors? They have already given us an advantage at the storage medium level: formatting became 10% effective, error correction – simpler. Just imagine, how much money has already been invested into infrastructure built around 512 bit sectors…
I would show you WD drive with native support for 4K right now (just have to deliver it from the USA first). As soon as the market would be ready for it we will start selling them immediately. One-two months preparation should be enough.
Helium-filled hard drives. Up to seven platters in enclosure.
Q: HGST has recently demonstrated a helium-filled hard drive. Will hard drives containing seven magnetic platters really be commercially available?
A.B.: Helium is good because it is less dense as compared to the air. Due to lesser density you can make magnetic platters lighter in weight and also you can install large platters in enclosure, up to seven pieces. Think of the companies that want to get maximum data density per square meter. They would love such drives.
Q: Isn’t it possible to simply discharge the air inside an enclosure?
A.B.: We all use aerodynamic technologies to control the distance between head and surface of a carrier, and I don’t know what would be the physics when using discharged air. I have also thought about using hydrogen instead of helium, but I can’t tell anything about it as well. It is easier to get hydrogen than helium – through water electrolysis. On the contrary, helium resource is limited and in order to get more we need tokamak – a fusion device. Although if there is enough helium to fill balloons, then, I suppose, there would be enough to fill hard drives.
In reality we already use helium during production. A drive is filled with helium during a stage that requires pinpoint accuracy in positioning heads. Difficulty in applying helium in commercial products is in keeping it inside at least during the warranty period. Gas molecule is too small and that is why it easily gets out through the finest pores. I am not aware of the details of how HGST has done it, but if you check the structure of hard drive, you will see elements that need a special attention to do it.
First thing is to block the air filter. A drive will turn into a pressurized vessel. Second thing is to make sure that the washer between the lid and enclosure is not porous for helium. This is as far as materials are concerned.
Q: Will there ever be helium-filled hard drives under WD brand?
A.B.: I never say never. We already use helium during production and we know the technologies required for that. In addition to that, the advantage in our industry is often measured by months, not by years.
Reliability, durability and statistics
Q: What is an estimated lifespan of the drive?Lets take a hard drive with MTBF (mean time between failures) at the level of 1 million hours. In terms of years it would give an unrealistic lifespan of 114 years. It looks as if MTBF simply cannot be converted into years…
A.B.: In fact MTBF is not an assertion about durability of the drive. It is a statistical assertion about drive population. Million hours of MTBF tells me that I, being a data centre manager with one hundred thousand drives can expect that one of them would fail every 10 hours. This will give me an understanding of how many spare drives I need to keep in stock.
MTBF is estimated with the help of the following test. If you need to get MTBF 2 million hours, you need to generate over 2 million work hours. Fill the racks with one thousand drives and keep them on during one thousand hours (six weeks). This is how we get a million hours. We can also increase a test stand temperature and by that streamlining drive amortization by 4 times. Now we have 4 million hours. If drives failed within this time, we get MTBF 2 million hours. In fact we started removing MTBF from device documentation because people often misinterpret its meaning.
Nevertheless, what is a service life of the hard drive? According to conservative estimate it is not less than 5 years. It is unlikely that you know a person who uses computer older than that. Whether it will live longer – depends on operation. Its like a car. If you drive a hundred kilometres a day it wears faster than a car that is taken from a garage only on a weekend.
It is difficult to give more accurate estimate since there only a few ten-year-old drives. Capacity of such drive is only 10-40 Gb and most of them are in computers that not used anymore. No data. All we have in this regard are private and sometimes absurd testimonies. For example, one of my clients has recently written off an array of WD drives that were purchased five years ago and all still were in working condition. And it was a 24/7 data centre.
Corporate hard drives are initially designed with a focus on reliability. It includes components selection and design. For example, 2,5-inch form factor hard drives with 15 000 RPM use platters with a capacity of 300 Gb. At the same time the largest capacity of a platter of this size is 500 Gb. Reliability is a serious discipline. Here, in Western Digital, this matter is dealt with by a lot of people with PhD.
An important factor that contributes to hard drive’s reliability is a temperature. Higher it is, more often they fail. Temperature is everything. We have a model of connection between the number of failures and so called effective activation energy into which temperature contributes as well. Seagate, HGST, Toshiba also have similar models. They are very precise since it is easy to measure and test it. Advice for data centre owners: keep your hard drives within the range of 40–50 °C.
Balancing magnetic platters
Q: One of the technologies that distinguish recently released WD Red series hard drives as well as corporate WD drives – 3D Active Balance Plus. What does it mean?
A.B.: It is like in the case with a car wheel, a stack of magnetic platters on spindle is influenced by disbalancing forces in two directions: one shakes (platters vector), another oscillates (perpendicular vector). In order to compensate the ‘shaking’ force, we put a piece of wire in the form of a discontinued ring in the motor mount. In order to deal with the ‘oscillating’ force we put a ‘cork’ in one of the holes on a spindle. A tiny coloured piece of plastic.
On destiny of former Hitachi GST
Q: Several month back Western Digital has acquired HGST (previously known as – Hitachi GST). What WD is going to do with such asset?Are you going to continue selling HGST devices under a separate brand? And finally, what is more valuable for WD in this deal: production capacities of HGST or its intellectual property?
A.B.: Answer to the question of what we are going to do with Hitachi: we won’t do anything with it. Such is a rule set by Chinese ministry of commerce. We have to keep two fully independent and competing companies during at least two years. There is main company WD Corporation and there are two fully independent affiliated companies WD and HGST. Here, in WD we are not allowed to talk with HGST neither about technologies, nor about prices, nothing at all! If you worked in Hitachi I would have to stand up and leave immediately.
When two years pass we can think of several scenarios of further developments. It may happen that two competing companies is a good option. They may be united and use the best of what they have. But I am an engineer and do not participate in decision making, that is the sphere of top-management. I can only speculate about it.
On R&D and competitive advantage
Q: How important is it to have an advantage in R&D to successfully compete on the market of hard drives? Or, perhaps, hard drives made by different companies are similar at large and you cannot make something innovative because similar things are simultaneously done by your competitors.
A.B.: That is an interesting question. Lets take Microsoft and its competitors. You can’t take one OS and replace it with another directly. Or Intel, AMD and ARM: their products are not directly interchangeable. It is all in other way with HDD because we have standards established by ATA Committee so that drives are interchangeable. And it gives rise to an intense competition. You are right, drives are 95% same and the difference between them constitutes only 5%. And, maybe, even 5% is too high estimate. Maybe it’s only 2%.
But I have to say that it’s great to be in a competitive business because competition keeps you fit and makes you develop and improve consumer products. We continue the race for achieving new record density, new capacity level, new technologies. It happens that one company is first to introduce a certain combination of properties, for example a product with SAS interface and 4 TB capacity, in which WD has a little advantage. However this leadership won’t last long, only 3-6 months. Such competitive advantage works when people need a certain product. It is important to release a product at the right time and under right market conditions. By and large the advantage could be little but in our case little means a lot.
How to start producing HDD in Moscow’s vicinity
Q: Would it be difficult to create a new HDD manufacturer?
A.B.: If we go back 30 years from now, we will see around 70 companies producing HDD. Now there are three or five, depending on how you count. These companies have absorbed the resources of all seventy predecessors. At the end of the 90s Seagate acquired Conner Peripherals, at the beginning of 2000s Maxtor acquired Quantum. Hard drive division of IBM, an original HDD inventor was acquired by HGST that in its turn was recently acquired by WD. And Seagate acquired Maxtor. It means, for example, Seagate today is a conglomerate of Maxtor, Quantum, Conner, Samsung and others.
Consolidation of hard drive manufacturers (photo from Wikimedia Commons)
Main thing here is an intellectual property, because those companies that are left on HDD market have amassed all IP required to produce hard drives. There are agreements among companies on distribution of some technologies without which you cannot make a hard drive. But its not the same situation as with Intel and AMD. The latter I would describe as more artificial.
If you and me win one million Euro in a lottery and decide ‘hey, I want to establish HDD production’, despite all our money we won’t be able to do that because we won’t have a technology rights. Seagate, WD and Toshiba would chase us in a court, and our company would go down. Like a lead airship!
From production perspective it is easy to start making bad drives. We could together make a rudimentary drive from radio market components. But making it is one thing, other thing is making it with the latest technologies, tiniest size of components, maximum record density – it is indeed difficult. We have years, decades of experience. Every day there is a new initiative introduced in the factory, to improve something. This business has a very rapid curve of training and high entrance barrier.
What, hard drives haven’t died yet?
Q: What WD thinks about a threat from SSD in the short and long range? Will hard drives survive during the next ten to twenty years or will they be forced out completely by solid state drives?
A.B.: Once in several years someone says: hard drive is dead and another technology will replace it. This is a very stable trend. But let me give you a nind experiment. Once again imagine that hard drives have never been invented. So we sit here at the table and I say: I have 128 Gb capacity on SSD on laptop and I need more. Who would have thought that we can make a spinning magnetic disk and wirting head will fly two nanometers above its surface? We wouldn’t believe that its even possible. However, we have it here and now!
Hard drives: 8 to 1-inch form factor (photo from Wikimedia Commons)
Important aspect: gorwth rates of specific capacity per dollar for SSD and HDD are parallel. Every step of HDD upwards is reflected in SSD but curves remain parallel.
In addition, there is noe problem that poses a challenge to SSD manufacturers, including WD, because we also have SSD department. Only a limited number of write operations is admissible – then the drive will become useless. It might be 100 thousand, or 10 thousand. But as the size of SSD component reduces, its lifespan becomes shorter. On the one hand, you increase the capacity, on the other you reduce its life. Hard drives do not lose an ability to read or write. You can read or write after five years same way you did at the beginning. We don’t have any restrictions.
Q: But sooner or later we will face a minimum cell size in NAND memory…
A.B.: If we think about theoretical minimum size of a cell in NAND memory, it will equal to one electron moving from a substrate into a floating caisson and back. Ideal storage for one bit. Possible problems with reliability are solved with the help of redundancy, error correction.
Q: What is theoretical limit of hard drive capacity?
In mid-1990s there was a dialogue about maximum capacity of a hard drive at the rate of 100 Gbit/sq.in – so-called superparamagnetic limit. It is not possible to magnetize something over it so that it remains magnetized. Today we produce products with record density of 720 Gbit/sq.in – safely and in large quantities. Those, who were saying that superparamagnetic limit restricts density to 100 Gbit/sq.in turned out to have been wrong. What is the limit now? I really don’t know. Maybe it’s a single electron whose spin we will manipulate.
A hard drive will still be with us for another several decades. Our technological plans already go up to 2020. SSD and HDD perfectly coexist in the structure of a multi-level storage. Lets take it wider: we have a static memory in processor cache at the top – very expensive and very fast. Then DRAM, then SSD, hard drives, tape. We have a continuum “performance vs. price”.
Western Digital’s hybrid HDD
Here also penetrate hybrids, something between SSD and HDD, trying to combine advantages of both technologies. We have already announced a hybrid hard drive 5 mm thick designed to suit the needs of ultra notebook market. We like the idea of introducing hard drives to places where previously only SSD has been used. A hybrid will allow getting 500 Gb capacity with performance almost like SSD and almost at a hard drive’s price.
Compared to Seagate’s Momentus XT it works slightly differently. It uses a driver – an intellect that allows determining where will data go inside a storage device: to SSD, to hard drive or both. It determines a suitable place for every piece of data on the basis of access history, statistics of operations. This device has a standard SATA interface and solid state and disk components are connected inside with a ‘bridge’. Therefore, operating system sees this drive as an ordinary hard drive but the driver knows about existence of a bridge and can access it without letting OS know.
Q: In contrast, Momentus XT works without knowing what happens above block level. Is this approach worse than yours?
A.B.: It also has advantages. Its simple and you don’t have to install the driver. Especially when there is simply no driver available for a certain operating system.
MLC is used as NAND memory. In our prototypes we tried SLC memory as well but it turned out that MLC perfectly fits to this product and in addition it has an advantage of SLC in terms of price.
Check how tightly we pack electronics in our drive. And here is a compact form factor connector. It is standardized (SFF-8038). The enclosure is thin, therefore it is important that one cannot force through its middle. For that purpose a spindle is tied with screws from two sides, like a bycicle wheel. Ordinary hard drives have it like a car wheel. Spinning speed – 5400 RPM. 7200 RPM would be too much since performance is increased due to flash memory.
Q: Is WD going to release a 3.5-inch form factor hybrid drive?
A.B.: At present we plan a version of this disk 7 mm thick with two platters and same way of arranging electronics. As far as 3,5-inch hybrids, we do not announce such products at the moment, but you may expect it sooner or later. Lets put the question this way: is a demand for such technology is convincing enough? Now there is a demand in something thin, something that suits ultrabooks.
One of my specializations is consumer electronics. Buyers want a hybrid drive. I ask them, why they need it. Then they look at their legs and start studying them. So, it turns out that they don't know why they want it!
Q: After what you have said it seems that hard drive is something like a Swiss watch. Such complex, elegant, ‘old-fashioned’ in its own way technology…
A.B.: Not bad. But watch is a rough and clumsy item compared to disk drive. We operate at a scale of nanometres. Hard drive is 3-4 times more complex than watch. It is just an offence: he compared us with Swiss watches!