Key Points
- A London startup, Vaire Computing, has created a new energy-saving chip called Ice River.
- The chip is designed to recycle and reuse some of the energy it consumes. It uses two key techniques: “reversible logic” and “adiabatic computing.”
- In a proof-of-concept test, the chip used 30% less energy than a standard processor.
- The technology aims to address the rising energy demands of AI systems.
A London-based startup has created an experimental computer chip that can recycle and reuse some of the energy it consumes, a potential breakthrough in the fight against the massive power demands of artificial intelligence. The company, Vaire Computing, hopes its “Ice River” chip will help make the next generation of AI systems much more energy-efficient.
In a recent test, the Ice River chip used about 30% less energy than a standard processor while performing the same task. It achieves this by tackling two major sources of inefficiency in modern chips.
First, it utilizes “reversible logic,” which enables the chip to perform calculations in both directions. Instead of using energy once and discarding it as heat, this design allows the chip to reuse its inputs for further calculations.
Second, the chip employs a technique known as “adiabatic computing.” In a normal chip, voltages change abruptly, like a hammer striking down, which wastes a lot of energy as heat. The Ice River chip, however, changes its voltages gradually, more like a gentle pendulum swinging back and forth. This “sloshing” of energy allows the system to recycle power for its subsequent operation.
As one of the company’s scientists explained, current devices “use energy once and then throw it away.” This new design represents a major shift toward a more subtle and efficient approach to computing.
Although still in its early stages, the technology has already progressed beyond the proof-of-concept phase. Vaire Computing is now focused on long-term development, having recently hired a former technology expert from chip designer Arm and joined a tech incubator to support its work.
