AI Accelerators

AI is many things, but power efficient is not one of them. When AI is working, especially doing learning and inferencing, this requires extreme computational horsepower.

AI accelerator cards have made their marks with graphics processing units (GPUs), multicore scalar processors (CPU), and spatial accelerators, such as the tensor processing unit (TPU). GPU and TPU processing units offer significant computation speed by leveraging parallel processing cores, which help accelerate and enable DL for training and inference. However, a relatively large power consumption is also required for the computation and data movement. That’s because AI requires much higher processor utilization, and processors—especially GPUs—are power hungry.

When building infrastructure that contains an AI accelerator, designers must consider power delivery, thermal requirements, and form factor constraints. ADI combines the key advantages of high efficiency, low heat/power dissipation, and a scalable output power delivery, all in the smallest possible total solution size. In short, ADI delivers best-in-class transient performance with 40% reduced output capacitance.

ADI’s power management products are differentiated by the following technological advantages:

• Monolithic Integration: An optimized power process technology that mitigates or nearly eliminates parasitic resistance and inductance between the FETs and drivers, enabling high switching speeds with minimal loss.
• Coupled Inductor Technology: ADI’s patented technology enables higher density, larger bandwidth, faster transient solutions, 50% higher efficiency, and 1.82× smaller magnetics. 
• SilentMOS™ technology (Silent Switcher® technology): Faster switches with near 0 ns dead times. This architecture is designed to minimize EMI/EMC emissions while delivering high efficiency at frequencies up to 3 MHz.
• Highly integrated multiphase controller and power stage ICs: Smart power stage (SilentMOS) IC with integrated current and temperature sensors.
• Advanced Packaging: Best-in-class thermal performance with dual-side cooling in our high density design.