The new chip is 0.7 nanometers, smaller than IBM's two-nanometer chip that was first unveiled in 2021. But the circuitry of the new chip has been significantly adjusted. That older, larger process laid the transistors flat in what IBM Research called nanosheets, as CNET reported in 2021.
Now, the new 0.7nm chip uses IBM's recently developed nanostack architecture, which stacks the nanosheets vertically.
IBM says the new architecture results in better performance. In the company's experiments, it found the new chip improved performance by up to 50% and energy efficiency by 70% compared to the 2nm version.
According to IBM, the nanostack architecture also allows for a 40% smaller die for SRAM -- static RAM is a type of memory that doesn't require a constant flow of electricity to store data, and because it's faster than DRAM, it's in high demand for AI applications.
The chip won't be ready to go for a while, though. IBM is still working with its manufacturing partner, Rapidus, a Japanese foundry (chip-making factory), to ramp up. IBM says it "sees a path to production" in five years, but the demand for energy-efficient computing hardware is only growing.
Chips like those designed by IBM, Nvidia, AMD and others are the backbone of the AI industry. As AI developers like OpenAI and Google race to build the most advanced models, they need massive amounts of energy, or compute, to train them. But that can take a lot of electricity, clean water and land to devote to data centers.