Alex Daley, Chief Technology Investment Strategist

Do you have a smartphone?

Does it already do pretty much everything you’d like it to?

Now suppose you could upgrade to a “phone” that had a storage capacity of 100 terabytes—enough to record virtually your entire life—and a week’s battery life per charge: would you buy it?

You don’t have to decide right this minute, because such a device is not here yet. But it’s probably coming in the not-too-distant future. Because of memristors.

That’s a photo of Hewlett Packard CTO Martin Fink as he presented the object of his company’s #1 research project to an audience back in June. This announcement sent shock waves through the computer industry. HP has decided to roll the dice big time.

So what is that little square gizmo in Fink’s hands, and how does HP propose to make use of it? Simple. It’s a memristor-based chip, and with it HP intends to do nothing less than revolutionize the technology of computing.

By way of background, the development of computers over the past half-century was all made possible by the invention of the integrated circuit (IC). This is the very first one:

Courtesy of Texas Instruments

It was fabricated at Texas Instruments back in 1958 and is now preserved in its museum. With the advent of this humble-looking device, computing vaulted from the Stone Age of vacuum tubes into the modern era of semiconductors.

Since then, of course, the IC has been miniaturized and micro-miniaturized, until several billion can be crammed onto one little silicon chip. All the while, though, the architecture of a computer has barely changed.

Its basic building block remains the IC, or transistor, which for computing purposes is really nothing more than a switch. It has three terminals: the source, drain, and gate electrodes. The gate controls the electron density in the central region of the transistor, which is made of a semiconducting material. If the electron density is high, current flows from the source to the drain; it doesn’t if the electron density is low (i.e., the switch is either off or it’s on). Assign the two states values of “0” and “1” and you have the beginning of a binary computer on which the most complex of calculations may be performed.

That design has worked very well so far. Computing power has expanded roughly in accordance with Moore’s law—first advanced by the cofounder of Intel in 1965—which holds that the number of transistors that can fit on a given computer chip, per unit cost, will double approximately every two years.

But you cannot keep shrinking the size of a transistor indefinitely. At some point they get too small to properly perform their function. And that point is rapidly approaching. When it’s reached, assuming that we want computers to keep getting evermore powerful, then a new structure will have to be implemented.

Around the world, countless scientists and engineers are devoting their careers to coming up with that replacement. There are lots of candidates, with varying degrees of functionality. Some are working models, others merely promising ideas. They include: quantum computers; DNA computers; 3D chips; spintronics-based chips; graphene transistors; and many more. (We took an in-depth look at each of them in the latest issue of Casey Extraordinary Technology, for those interested in a deep dive.)

HP has cast its lot with memristors.

Memristors were first described in 1971 by UC Berkeley Professor Leon Chua. Chua posited that the then-current model of a logic circuit—composed of an inductor, resistor, and capacitor—was incomplete. He suggested a theoretical fourth building block for the circuit—one that has properties that cannot be duplicated by any combination of the other three elements: essentially, it would be a component that could remember what kind of charge last passed through it. Chua called it the memristor.

But no one had ever seen one. And that was the end of that for the next 30 years. Memristors’ existence still had yet to be proven when researchers at HP initiated a new project in the early 2000s. They weren’t looking for the elusive fourth building block, just trying to build a fast, low-power switch by placing two microscopic resistors in a stack and using the current in one to flip the resistance in the other. They succeeded in building the device they wanted, but couldn’t seem to predict how it would actually behave. After several years of fooling with it, they finally discovered the reason buried deep in the annals of computing research: they needed the idea of the memristor to complete the logical description of the machine.

Digging into Chua’s work, they discovered they had accidentally built the seemingly mythical device. In theory, they could now build a more complete computer.

In 2008, they turned theory into reality by creating a physical example: a circuit using memristors just a few atoms thick. The benefits of a memristor-based transistor quickly became clear. It was far faster than any kind of computer yet devised and used far less power. That’s because the memristor is fast enough to replace the incredibly expensive and small cache layer of a computer, yet has the capacity potential of a hard drive. No more slow and inefficient triple layers of memory—from the die to the RAM to the disk, all on one chip.

Those changes alone were promising enough, but in addition, it had some extraordinary side benefits: in the event of power loss, it maintained its memory. Rip a cord from a wall or battery from a phone, and nothing is lost. Plus it switched on and off as quickly as an electric light.

All of these—speed gain, memory retention, low power consumption, instant on/off—were obviously game changers. For the faltering company that’s shed tens of thousands of jobs in the past few years as it missed out on the smartphone and tablet revolutions, leaping from laggard to untouchably far ahead of the competition in one fell move is immensely attractive.

So the company decided that memristors were the wave of the future and that it would go all in. At the June press conference, HP announced that it was developing something called simply “The Machine,” a memristor-based computer projected for commercial development by 2020.

In the words of CEO Meg Whitman: “HP has been talking about the individual component technologies for some time and now we are bringing them together into a single project to make a revolutionary new computer architecture .... This changes everything.”

Perhaps it does. Perhaps HP will leave all of its competitors in the dust…

Provided it doesn’t kill its own golden goose.

It’s done that many times before. For example, the HP of old was one of the first players in the smartphone game. It inherited the first iPAQ (that name sound vaguely familiar?) mobile phone and PDA combo when it purchased rival computer-maker Compaq in 2002. For five long years before the first iPhone ever came to market, HP had that opportunity in the bag. Then it lost out to BlackBerry, iPhone, and Android in series before finally giving up on the market entirely.

2003 HP iPAQ—no lines outside the store for this one

HP also led most of its competitors to market in tablets, too. It released one of the first tablet PC notebooks, a series of ill-fated hybrid tablet laptops much like the new models all clawing at but failing to make any real dent in the iPad phenomenon. HP had nearly a decade lead in that market, only to relinquish it entirely in the end.

HP’s first tablet PC—not that dated even after a decade

But today we have an all-new HP under rock-star CEO Meg Whitman, formerly of eBay. In fact, the company is so bold about its new plans that it’s actually splitting in two.

The spinout strategy, reminiscent of Sears’ move to shed the Sears Hometown and L.L.Bean franchises it once gobbled up, is meant to make each half of HP leaner and more focused, according to management. There is no reason, of course, for them to duplicate all those HR, legal, and other centralized expenses—two businesses can easily live under one roof as Microsoft and GE and Siemens, and many others, have shown.

No, the real reason for the split is that the computer-manufacturing business is a drag on the value of the stock.

The first half of the split, the retail-focused HP, Inc., is to make personal computers, tablets, phones, and HP’s real legacy: printers.

Unfortunately, over the last two years HP’s laptop sales have fallen $5 billion. Printers and desktops have also each dropped another $2.5 billion or so. The company is losing market share to enterprise competitors like Lenovo—and it missed out on the shift to smartphones and tablets on both sides of the market, as well as losing out in retail presence to more forward-thinking companies like Apple with its eponymous stores.

Still, thanks to the one commodity with a seemingly never-ending price elasticity—printer ink—the division made $3.9 billion in operating profits last year and is expected to post $5.4 billion this year on $57.2 billion in revenue. So the company has the cash to tackle its missteps head-on… if it hurries.

But ironically, it won’t be that division which takes on the effort to build The Machine at all, despite the obvious mobile device benefits. That job will be left to Hewlett-Packard Enterprise (HPE), the company that will remain when HP, Inc.’s hardware division is spun off into its own company.

HPE will hang on to the company’s “big iron” products like Moonshot and Gen 9 servers that are sold only to massive companies and governments. And it will be home to the monstrous systems integration and consulting businesses (i.e., services)—HP’s answer to IBM. Together they are expected to produce $58.4 billion in revenue and $6 billion in operating profits.

HPE will also maintain the Labs division, of which supposedly 75% of the employees are focused on the multiyear effort to bring The Machine to market.

Although it might not seem to make sense at first blush, giving the game-changing mobile technology to the company without a mobile division actually makes sense. That’s because the energy savings, larger storage capacity, and higher throughput of the hardware is far more important to HP in another division: servers.

The largest customers of HP’s enterprise group are the big companies running huge server farms. There the largest cost beyond computer hardware itself is often electricity—for running the systems and cooling the monster boxes. Last time Google revealed numbers in 2011, it reported using 260 million watts of electricity a year, or the same amount required to power 200,000 homes.

Right behind that usually comes the cost of storing all that data, thanks to the massive arrays of individual spinning disks and solid-state drives required to serve up data fast enough for many millions of users. The many layers of controllers and cabling between the computer’s core and all those disks are just bottlenecks slowing down the flow.

The Machine theoretically solves all of those issues in one fell swoop. Whole chunks of the computer are ripped out and replaced with a single low-power board like the one shown at the top, allowing server farms to pack dozens of times more densely than today while running on much lower power.

And those customers have the cash to pony up big premiums for early versions of the hardware long before it’s refined enough to fit in a few-millimeters-thick space between your phone’s battery and screen.

HP is right to call memristor technology a game changer. And it’s probably right to keep it in its Enterprise half of the new company, too. Let’s just all hope for the sake of their shareholders they don’t fumble this one like they have in so many other markets.

Click to Comment on This Article