Transistor Mimics Human Brain’s Ability to Change Physical Structure

Supercomputers are, perhaps, humanity’s most iconic symbol of computational power. They can perform amazing feats of calculation, yet they are utterly gargantuan, consume enormous amounts of energy (typically on the scale of mega Watts) and are ineffective learners.

By comparison, the human brain is a biological marvel. With over 86 million neurons and about 100 trillion synaptic connections, the brain quickly learns by selectively strengthening or weakening connections to adapt to changes in stimuli. It also consumes only around 20 W of energy – about the same as a standard CFL bulb.

Researchers at the Harvard School of Engineering and Applied Science (SEAS) recently developed a new proof-of-concept transistor based on the brain’s ability to learn. The key is utilizing plasticity, which is the brain’s ability to change its physical structure based on mental activity.

The transistor is composed of a samarium nickelate semiconductor sandwiched between two platinum electrodes with a nearby pouch of ionic liquid to retain ions.

The device uses a multiplexer to convert the time delay of an electrical input signal into a voltage. The voltage is then applied to the ionic liquid, creating an electric field capable of forcing oxygen ions to move in and out of the nickelate thin film.

The result is a change in the overall concentration of ions in the nickelate, which also changes the conductance of the nickelate. This causes either an increase or decrease in the device’s ability to carry information via electrical current.

“The transistor we’ve demonstrated is really an analog to the synapse in our brains,” said researcher Jian Shi. “Each time a neuron initiates an action and another neuron reacts, the synapse between them increases the strength of its connection. And the faster the neurons spike each time, the stronger the synaptic connection. Essentially, it memorizes the action between the neurons.”

The samarium nickelate used in the transistor is part of a special class of materials called correlated electron systems, which can undergo an insulator-metal transition under special circumstances. In this case, the conductance of the nickelate changes when exposed to an electric field, causing ions to move into or away from the material. Because it’s extremely sensitive, the energy input to produce a large signal is quite low, so the material is intrinsically energy efficient.

According to Shi, “[the device] changes its conductance in an analog way, continuously, as the composition of the material changes. It would be rather challenging to use CMOS, the traditional circuit technology, to imitate a synapse, because real biological synapses have a practically unlimited number of possible states, not just “on” or “off”.”

The new synaptic transistor unleashes many exciting possibilities for computing, as the device isn’t restricted to a binary system and offers considerable opportunities for energy savings. It can also operate in environments up to at least 160°C, which would fry a biological brain.

“There’s extraordinary interest in building energy-efficient electronics these days,” said principal researcher Shriram Ramanathan. “Historically, people have been focussed on speed, but with speed comes the penalty of power dissipation. With electronics becoming more and more powerful and ubiquitous, you could have a huge impact by cutting down the amount of energy they consume.”

The prototype device is embedded in a silicon chip, so the researchers believe it will integrate well with existing silicon-based systems. Additionally, it has non-volatile memory, meaning that it’s able to remember its last state despite power interruptions.

Naturally, there are limitations with the technology in its current state. Because the nickelate is a relatively new class of material, a limited amount of research is available on its synthesis.

“You have to build new instrumentation to be able to synthesize these new materials,” said Ramanathan, “but once you’re able to do that, you really have a completely new materials system whose properties are virtually unexplored.”

Controlling the device size is also imperative, as it affects the speed of the transistor. According to Ramanathan, the time constant in the prototype device was defined by the experimental geometry. “To make a super-fast device, all you’d have to do is confine the liquid and position the gate electrode closer to it.”

“This kind of proof-of-concept demonstration carries that work into the ‘applied’ world, where you can really translate these exotic electronic properties into compelling, state-of-the-art devices.”

Transistors with the ability to learn bring us one step closer to accessing the calibre of artificial intelligence that previously existed only in science fiction. Exciting? Very. Just don’t be surprised if our robot overlords show up sooner than you thought.

Images: Harvard SEAS