So let’s prep for theatre, scrub our hands clean, and dig into the anatomy of what we use today to hold onto our trillions of digital bits. We’ve split the Anatomy of a Storage Drive in three parts, all published at the same time to dissect hard disk drives, solid state storage, and optical drives. Follow the links below to read them all, along with our previous published work on the series.
You spin me right round, baby
Let’s start our look into the guts of storage drives with ones that use magnetism to store digital data. The mechanical hard disk drive (HDD) has been the standard storage system for PCs across the world for over 30 years, but the technology behind it all is much older than that.
Anatomy of a Motherboard Anatomy of a Power Supply (PSU) Anatomy of a Hard Disk Drive Anatomy of a Solid State Drive Anatomy of an Optical Drive Anatomy of a Graphics Card Anatomy of a CPU Anatomy of RAM Anatomy of a Monitor Anatomy of a Mouse Anatomy of a Keyboard Anatomy of a Gamepad
IBM released the first commercially available HDD in 1956, all 3.75 MB of it. And generally speaking, the overall structure hasn’t changed a great deal in that time. There are still disks, that use magnetism to store data, and there are devices to read/write that data. What has changed, and hugely so, is the amount of data that can be stored on them. Back in 1987, you could buy a 20 MB HDD for around $350; today that kind of money will get you 14 TB of storage: 700,000 times more space. We’re going to take apart something that’s not quite that size, but still pretty decent today: a 3.5" Seagate Barracuda 3 TB HDD, specifically the ST3000DM001 model, infamous for it’s high failure rate and subsequent lawsuits. This one is dead, too, so in truth this is more of an autopsy, rather than an anatomy lesson.
The bulk of the hard drive is cast metal. The forces inside the device, when under heavy usage, can be pretty serious, so the use of thick metal stops the body from flexing and vibrating. Even tiny 1.8" HDDs use metal for the body, although they tend to be made from aluminum, rather than steel, as they’re designed to be as light as possible.
Flipping the drive over, we can see a circuit board and a bunch of connections. The one at the top of the board is for the motor that spins the disks, whereas the bottom three are, from left to right, jumper pins to allow the drive to be configured for certain setups, SATA (Serial ATA) data, and SATA power.
Serial ATA first appeared in 2000, and in desktop PCs, it’s the standard system used to connect drives to the rest of the computer. The specification of the format has gone through lots of revisions since then and we’re currently on version 3.4. Our hard drive cadaver, though, is an older version but this only affects a single pin in the power connection. The data connections use what is called differential signalling to send and receive data: the A+ and A- pins are used to transmit instructions and data to the hard drive, whereas the B pins are used to receive those signals. The use of paired wires like this greatly reduces the impact of electrical noise in the signal, which means it can be run faster. On the power side of things, you can see that there are essentially two of every voltage (+3.3, +5, and +12V); most aren’t used, though, as HDDs don’t need a lot of power. This particular Seagate model uses less than 10W under heavy load. The power pins labelled with PC are pre-charge ones: it allows the hard drive to be pulled in and out, while the computer is still on (a.k.a. hot swapping). The pin labelled PWDIS allows for remote resetting of the hard drive, but this is only supported by SATA version 3.3; so in our drive, it’s just another +3.3V line. And the last pin to cover, the one marked SSU, just tells the computer as to whether or not the hard drive supports staggered spin up. The disks inside the device – which we’ll see in a few moments – must be spun up to full speed before the computer can start using it, but if the machine had lots of hard drives, the sudden simultaneous demand for power might upset the system. Staggering the spin ups helps prevent such problems ever occurring, but it does mean that you’d need to wait a few second more before you can get all jiggy with the HDD.
Removing the circuit board reveals how the circuit board connects to the components inside the drive unit. HDDs aren’t air tight, except for the super large capacity ones – these use helium, instead of air, as it much less dense and creates fewer problems for drives with lots of disks. But you don’t want them openly exposed to the environment either. By using connectors like this, it helps minimize the amount of entry points that dirt and dust can work their way into the drive; there is hole in the metal case - bottom left of the above image (big white dot) - to allow air pressure to remain relatively ambient.
Now that the circuit board is off, let’s have a look at what’s here. There are 4 main chips to focus on:
LSI B64002: the main controller chip that handles the instructions, data flow in and out, error correcting, etc. Samsung K4T51163QJ: 64 MB of DDR2 SDRAM, clocked at 800 MHz, used to cache data Smooth MCKXL: controls the motor that spins the disks Winbond 25Q40BWS05: 500 kB of Serial Flash memory, used to store the drive’s firmware (bit like a PC’s BIOS)
There’s little difference across the vast range of HDDs out there, when it comes to the components on the circuit board. Larger storage requires more cache (you can find up to 256 MB of DDR3 on the latest monsters) and the main controller chip might be a little more sophisticated with regards to handling errors, but there’s not much in it.
Opening up the drive is easy enough, just unscrew a bunch of Torx fittings and voila! We’re in…
Given that it takes up most of the unit, our attention is immediately drawn to the big metal circle, so it’s not hard to see why they’re called disk drives. The proper name for them is a platter and they are made of glass or aluminum, coated with multiple layers of different compounds. This 3 TB drive has three platters, so each one must store 500 GB on each side.
The image of these dusty, hairy platters doesn’t do any justice to the engineering and manufacturing precision required to produce them. In our HDD example, the aluminum disk itself is 0.04 inches (1 mm) thick, but it has been polished to such a degree that the average height of the variations in the surface is less than 0.000001 inches (roughly 30 nm). A base layer of just 0.0004 inches (10 microns) deep, comprising several layers of compounds, has been applied to the metal. This is done through electroless plating and then vapor deposition, which preps the disk for the crucial magnetic material that’s used to store the digital data. This material is usually a complex alloy of Cobalt and is set out in concentric rings, with each one being around 0.00001 inches (roughly 250 nm) wide and 0.000001 inches (25 nm) deep. On the microscopic scale, metal alloys form grains, like soap bubbles floating on water. Each grain has its own magnetic field, but it can be aligned into a set direction. The grouping of these fields gives rise to the 0 and 1 bits of data. If you want a deeper technical dive into this topic, have a read of this document from Yale University. The final coatings are a layer of carbon for protection and then a polymer to reduce contact friction. Together, they come to no more than 0.0000005 inches (12 nm) thick. We’ll see why the platters have to be made to such high tolerances in a moment, but it is astonishing to think that, for as little as $15, you can be the proud owner of nanometre-scale manufacturing! Let’s go back to the whole HDD again, and have a look at what else is there.
The yellow box highlights a metal cap that holds the platter firmly in place on the spindle motor – the electric drive that rotates the disks. In this HDD, they rotate at 7200 rpm, but other models run slower. Slower drives keep noise and energy consumption down, but also lowers performance, while other faster drives can reach 15,000 rpm. To help reduce the damaging effects of dust and moisture in the air, a re-circulation filter (green box) picks up tiny particles and traps them inside. The air moved by the rotation of the platters ensures there is a constant flow over the filter. On top of the disks, and next to the filter, is one of three platter separators: these help reduce vibrations and also keep the air flow as regulated as possible. In the top left of the image, indicated by a blue box, is one of two permanent bar magnets. These provide the magnetic field that is needed to move the component highlight in red. Let’s clear out some of these parts to see this better.
What looks like a chunky Band Aid is another filter, except this one clears out particles and gases from the outside, as they enter through the hole we see before. The metal spikes are actuator arms that hold the hard drive read/write heads - they sweep back and forth across the surface of the platters (top and bottom) at a ridiculously high speed.
Watch this video courtesy of The Slow Mo Guys to see just how fast it is: Rather than use something like a stepper motor, to snap the arms into place, an electric current is sent around a coil of wire at the base of the arm.
These are generally called voice coils, because it’s the same principle that loudspeakers and microphones use to move the soft cones. The current generates a magnetic field around it, which reacts against the field made by the permanent bar magnets. Don’t forget that the data tracks are tiny, so the positioning of the arms needs to be extremely accurate – just like everything else in the drive. Some hard drives have multi-stage actuators, that can do smaller changes in direction with just part of the whole arm. On certain hard drives, the data tracks actually overlap each other. That technology is called shingled magnetic recording, and the requirement for accuracy and precision (i.e. hitting the right position over and over) is even greater.
At the very ends of the arms are the delicate read/write heads. Our HDD has 3 platters and 6 heads, and each one floats above the disk as it spins. To be able to do this, the heads are suspended by two ultra thin strips of metal. It’s here that we can see why our anatomy sample is dead – at least one head has come loose and whatever caused the original damage, also bent some of the support arms. The whole head component is so small, that it’s really difficult to get a good image with a regular camera, as we can see below.
We can make out some parts though. The grey block is a specifically machined part called a slider, as the disk rotates underneath it, the flow of air produces lift, raising the head off the surface. And when we say “off,” we’re talking about a clearance of just 0.0000002 inches or less than 5 nm. Any further away and the heads wouldn’t be able to detect the changes in the magnetic fields in the track; if the heads actually rested on the surface, they would just scrape off the coating. This is why the air inside the drive case needs to be filtered: dust and moisture on the disk surface would just wreck the heads. The tiny metal ‘pole’ at the end of the head is there to help with the overall aerodynamics. We need a better picture, though, to see the parts that do the actual reading and writing.
Image: Roman Starkov, Wikimedia Commons In the above image, from a different hard drive, the parts that read and write are underneath all the electrical traces. Writing is done with an thin film induction (TFI) system, whereas reading is done with a tunneling magnetoresistive (TMR) device.
The signals produced by the TMR are very weak and have to be run through an amplifier, to boost the levels, before they can be sent onward. The chip responsible for this can be seen near the base of the actuator arms, in the image below.
As mentioned in the introduction of this article, the mechanical components and operation of a hard disk drive haven’t changed a lot over the years. It’s the technology behind the magnetic track and the read/write heads that has improved the most, producing narrower and denser tracks, which ultimately results in more storage capability. However, mechanical hard disk drives have clear performance limitations, It takes time for the actuator arms to move to the required position and if the data is scattered about in different tracks on separate platters, then the drive will spend a relatively large number of microseconds hunting down the bits. Before we move on to pull apart another type of storage drive, let’s make a reference point for the performance of a typical HDD. We’ve used CrystalDiskMark to benchmark a WD 3.5" 5400 RPM 2 TB hard drive:
The first two rows display the number of MB per second throughput for doing sequential (a long, continuous list) and random (jumping about the disk drive) reads and writes. The next row, shows an IOPS value, that’s the number of input/output operations taking place each second. The last row displays the average latency (time in microseconds) between the read/write operation being issued and the data value being retrieved. Generally speaking, you want the values in the first 3 rows to be as large as possible, and the last row to be as small as it can be. Don’t worry about the numbers themselves, it’s just something we’ll use for comparison, once we look at the next type of drive: solid state storage. Continue reading the Anatomy of SSDs here. Masthead credit: Patrick Lindenberg
