In the half-century history of electronics trade magazines, there are only a handful worth collecting as landmarks — the issue of Popular Electronics that announced the first personal computer, the “Would You Sell a Computer to Hitler?” issue of Computer Decisions, Upside‘s “Has Silicon Valley Gone Pussy?”, Wired‘s Sacred Heart Apple cover and a few others. This month’s issue of EETimes may be the first to make that exclusive list in this century. The reason? It announces a new kind of circuit that may, in fact, change the world.
Being a nice sober trade magazine dedicated to the technology behind the technology of the semiconductor world, EETimes is not usually driven to superlatives, but the August 18th cover — “Will Memristors Be Irresistible?” — suggests a barely constrained excitement that explodes on the pages inside.
Complete with a timeline that links this new technology with all of the great milestones in the history of semiconductors beginning with the transistor, EETimes then breathlessly suggests that the memrister — the missing fourth leg in the world of circuits — could soon be as big as its three (resistor, capacitor and inductor) multi-billion dollar predecessors. The article goes so far as to suggest that memristors could not only replace the hottest chip business on the planet, flash memory, but also play a crucial role in the development of adaptive control circuits “that learn” — i.e., the electronic synapses needed to create true ‘thinking computers.’
Pretty heady stuff. So why haven’t we heard about these memristors before?
One answer is that they are comparatively new. Whereas the other three circuit types date back to the crystal radio and instrument eras of the early 20th century, it wasn’t until 1971 that Leon Chua of UC Berkeley proposed just such a circuit, and not until 1980 did he publish a model for its design. Chua, a formidable figure in the circuitry world (he is also the father of neural networks), proposed the memristor based upon the simple, but brilliant, notion that given the various combinations of passive circuits that compare two features of an electronic current — resistors measure voltage to current, capacitors measure voltage to charge, and inductors measure flux to current — one obvious combination was missing: comparing flux to charge.
What made this missing combination particularly interesting, Chua realized, was the resulting device would be analog, not digital, and would ‘remember’ how much current has flowed through it. In other words, it would behave a lot like a nerve cell.
Now, roll forward to the mid-1990s. The semiconductor industry, thanks to the rapid pace of Moore’s Law, is beginning to worry about the eventual physical limitations of silicon as medium for future chips, especially as features dip down to near-atomic levels. Dozens of initiatives are undertaken at universities, government labs, and corporate R&D departments, all searching for potential alternatives — from quantum dots, to new non-silicon substrates, to new transistor architecture. The result is that, over the last five years, we’ve heard one announcement after another emerging from various labs announces breakthroughs in semiconductor design.
But one of these iniatives that managed to stay below the media radar was undertaken at Hewlett-Packard — and may prove to be CEO Lew Platt’s greatest legacy. In 1995, HP announced the creation of a“Quantum Structures Research Initiative”, to be led by chemist Stanley Williams and chartered to come up with molecular-level alternatives to the transistors. The fact that Williams was a chemist proved to be critical: many labs have investigated similar technologies in recent years, but it appears that only the HP team managed to come up with an actual physical design for a memristor by using titanium-dioxide nanostructures to create the ‘crossbar’ architectures of their experimental circuitry.
The exciting news in this is that the HP team now believes it can not only fabricate the memristor on a practical basis, but that the performance results of these circuits — called RRAMs — trump any other technology now available, including flash memory. We’re talking 100 gbits per square centimeter, versus 32 gbits for state-of-the art.
That’s the good news. One nagging concern is that when the first news of HP’s RRAMs leaked out, flash memory was only at 5 gbits per square centimeter. In other words, flash is rapidly closing the gap — and it wouldn’t be the first time that a new memory technology was rendered obsolete by an existing technology that stayed close enough that the high capital cost of switching became decisive. Experts say that for HP to be remain safely ahead, it will have to push RRAMs to 1 terabit per square centimeter.
For RRAMs to fail as a viable technology might have some far-reaching consequences. Memory isn’t the only potential application for these new circuits. Because RRAMs are capable of ‘remembering’ slight changes in resistance, they should be able to behave like nerve cells — opening the potential for massive memrister arrays that behave like human brains, learning an adapting as they go.
That scenario is still five years out, at best. In the meantime, for all of memristor technology’s promise, Hewlett-Packard still has to make it work in the real-world competitive arena. If HP can pull it off, memristors may indeed change the tech world forever — and send the protean Hewlett-Packard Co., itself as old as the electronics industry, through yet one more transformation as well.