Research teams at Intel Corp (INTC.O) on Saturday unveiled work that the company believes will help it keep speeding up and shrinking computing chips over the next ten years, with several technologies aimed at stacking parts of chips on top of each other.
Intel’s Research Components Group introduced the work in papers at an international conference being held in San Francisco. The Silicon Valley company is working to regain a lead in making the smallest, fastest chips that it has lost in recent years to rivals like Taiwan Semiconductor Manufacturing Co (2330.TW) and Samsung Electronics Co Ltd (005930.KS).
While Intel CEO Pat Gelsinger has laid out commercial plans aimed at regaining that lead by 2025, the research work unveiled Saturday gives a look into how Intel plans to compete beyond 2025.
One of the ways Intel is packing more computing power into chips by stacking up “tiles” or “chiplets” in three dimensions rather than making chips all as one two-dimension piece. Intel showed work Saturday that could allow for 10 times as many connections between stacked tiles, meaning that more complex tiles can be stacked on top of one another.
But perhaps the biggest advance showed Saturday was a research paper demonstrating a way to stack transistors – tiny switches that form the most basic building bocks of chips by representing the 1s and 0s of digital logic – on top of one another.
Intel believes the technology will yield a 30% to 50% increase in the number of transistors it can pack into a given area on a chip. Raising the number of transistors is the main reason chips have consistently gotten faster over the past 50 years.
“By stacking the devices directly on top of each other, we’re clearly saving area,” Paul Fischer, director and senior principal engineer of Intel’s Components Research Group said in an interview. “We’re reducing interconnect lengths and really saving energy, making this not only more cost efficient, but also better performing.”