Engineers led by Hong Kong Polytechnic University report a new kind of transistor meant to use less energy than the switches that now fill computer chips. Ordinary chips are built from MOSFETs, metal-oxide-semiconductor field-effect transistors. Those devices turn on when heat helps electric charges climb over an energy barrier. That process is called thermionic emission. At room temperature it cannot be sharpened past a floor known as the Boltzmann limit. The limit has stalled further cuts in chip power.
The PolyU device is a tunnelling field-effect transistor, or TFET. Instead of charges hopping over a barrier, they pass through it by quantum tunnelling, a process allowed by quantum mechanics when the barrier is thin enough.
The engineers said an industry roadmap already lists TFETs as the most promising substitute for MOSFETs, and that replacing thermionic emission with tunnelling is what lets this transistor cross the 60 millivolt-per-decade line.
How the layers were built
Earlier experimental TFETs often switched steeply but could not supply much current. The engineers stacked ultrathin sheets of bismuth and indium selenide, a heterostructure of two-dimensional materials, using pulsed laser deposition. In that method a laser pulse knocks material from a solid target so it can settle as a film. In this thin form, bismuth, usually a semimetal, behaves as a semiconductor. The layers line up so charges can tunnel efficiently into the indium selenide.
The Bi/InSe TFET kept a subthreshold swing below the Boltzmann floor across six orders of magnitude of current. It ran at room temperature on ordinary centimetre-scale silicon wafers. The gate needed only 160 millivolts of swing, against about 800 millivolts for advanced MOSFETs. Output current reached several microamps per micrometre, with a very high ratio between ON and OFF current. Strong output current matters because one transistor must be able to drive several others.
The same growth method is presented as a route to larger wafers and to very short transistor channels, and as compatible with silicon factories. The release frames the device as a building block for lower-power chips, including hardware for artificial intelligence (AI).
This research is published in Science.