It is also the most energy-intensive part of AI number crunching. The researchers call their new method Linear-Complexity Multiplication—it works by approximating FPMs using integer addition.