"Intelligent Transistor" Can be Programmed On-the-Fly for Improved Efficiency in AI Hardware
A team of scientists from TU Wien in Vienna have developed what they describe as an "intelligent transistor," a germanium-based device that could boost the energy efficiency of "emerging artificial intelligence electronics."
"We connect two electrodes with an extremely thin wire made of germanium, which is connected to metal on both sides with special, extremely clean interface," says Masiar Sistani, PhD and first author, of the device the team developed. "Above this germanium segment, we place a gate electrode like the ones found in conventional transistors. What is decisive is that our transistor also has another control electrode, which is placed on the interfaces between germanium and metal. It can dynamically program the function of the transistor."
That programmability is key: Where transistors in a traditional computer chip work in only one way, the "intelligent transistor" created by the team can be used to switch functionality on the fly — turning a NAND gate into a NOR gate, for example.
"Until now, the intelligence of electronics has come simply from the interconnection of several transistors, each of which had only a fairly primitive functionality. In the future, this intelligence can be transferred to the adaptability of the new transistor itself," explains Walter Weber, TU Wein professor and leader of the research group responsible. "Arithmetic operations, which previously required 160 transistors, are possible with 24 transistors due to this increased adaptability. In this way, the speed and energy efficiency of the circuits can also be significantly increased."
The device is to be of particular interest to those working in the field of artificial intelligent systems, the team claims. "Our human intelligence is based on dynamically changing circuits between nerve cells," Weber says. "With new adaptive transistors, it is now possible to change circuits directly on the chip in a targeted way."
"We don't want to completely replace the previous transistor technology with our new transistor, that would be presumptuous," adds Sistani. "The new technology is more likely to be incorporated into computer chips as an add-on in the future. For certain applications, it will simply be more energy-efficient to rely on adaptive transistors."
The team's work has been published under closed-access terms in the journal ACS Nano.