It’s difficult to evaluate what may be a fundamental advance in technology that happens to be mostly the product of a single company. The enthusiasm of one company doesn’t mean that the new technology – in this case the memristor (memory resistor) – will be accepted as fundamental to new products from other companies, in fact, it could mean just the opposite.
Hewlett-Packard (HP) didn’t invent the memristor, Leon Chua (University of California, Berkeley, USA) in a 1971 paper took some ideas that were floating around electrical engineering research and gave them a name and general theoretical base. However, it is HP that is most responsible for extending the concepts behind memristors into prototypes, for example, announcing in 2008 the development of a switching memristor with multiple possible applications including computer memory, computer logic, and neuromorphic (brain neuron mimicking) computers.
A Memristor – there is no such thing as the memristor, they vary for different applications – is constructed with nanoscale wires about 50 nanometers (150 atoms) wide. The wires are laid down in rows on a thin film. Then a sandwich layer of oxygen depleted titanium dioxide – the ‘business’ layer with the variable resistance properties – is spread over the nanowires. On top of that are laid down more rows of nanowire, perpendicular to the wires below. Each intersection of nanowire rows becomes a memristor. Applying a tiny current to a memristor can change the position of as little as one atom in the titanium oxide, which changes the resistance. That change is retained when the current is off. The difference in resistance can be ‘read’ by a current flowing in the opposite direction (using alternating current, AC) without changing the position. In this way, the memristor can store information, for example, as computer memory.
A big advantage for the memristor is the ability to pack tight rows, each memristor being as small as 3 nanometers, and then to be able to stack layers of memristors – a condensed format that can lead to high memory capacity in a very small space. Better yet, unlike traditional ‘flash memory’ (used in today’s memory sticks, etc.), which can be re-used about 100,000 times, the memristors can withstand about a million read-write cycles.
Arrays of memristors can be configured in ways that mimic the synapses of neurons (for example, in the brain), which has led to talk about simulating brain function with memristors. No less than Dr. Chua is quoted in a recent New York Times article…
Our brains are made of memristors,” he said, referring to the function of biological synapses. “We have the right stuff now to build real brains.”
[Source: New York Times]
The enthusiasm is understandable, and shared by quite a few people, but the context demands skepticism: We don’t know even know how brain memory works, much less the whole brain. It’s hard to see how a reasonable facsimile of the brain can be constructed on mostly supposition. Anyway, even the optimistic say that memristor technology won’t be advanced enough for such neuromorphic work until 2020-2025.
In the meantime, HP is preparing to take the memristor to market. That means ramping up for large scale production (sending it to the ‘fab’ in semiconductor parlance). The first target product is for computer memory (RAM) and storage. HP believes the memristor is reliably scalable and more efficient to manufacture than the current generation of silicon-based memory. There is no question that silicon technology is running toward the end of its capabilities. Many other approaches are being pursued. For example, IBM and Intel among others are experimenting with phase change memory, a system based on using heat to change the crystalline structure of a glass material.
All such technologies have to prove their worth in laboratory testing, which the memristor has done. However, they have to make the transition to mass production and general use. Therein lie conditions not envisaged, and many a product has failed to clear the hurdles. It will be several years before the memristor has run its course.