High efficiency power management IC adopted in AI products

Google Coral technology enables advanced neural network processing for low power devices. At the heart of Coral is the Google Edge Tensor Processing Unit (TPU) coprocessor, which is capable of performing four trillion operations per second (TOPS), using 0.5W for each TOPS (2 TOPS per watt). For example, it can execute advanced mobile vision models, such as MobileNet v2 at 400 FPS, in a power efficient manner.

The Renesas ISL91301B PMIC offers a high efficiency and small footprint power solution that is suitable for low power edge AI processors. The ISL91301B has quad buck regulator outputs, each delivering up to 4A (total 16A) with 94% peak efficiency. The PMIC leverages Renesas’ R5 Modulation Technology for quick single-cycle transient response, tuned compensation, and fast switching frequency.

These features and a 70mm² solution size make it easier for power supply designers to design a small AI solution with 2x2x1mm low profile inductors, small capacitors, and a few passive components. The ISL91301B’s low quiescent current, light load efficiency, high DC accuracy and fast dynamics is well suited for low power or battery-powered AI products.

“We are thrilled to be collaborating with Google on their innovative, next generation products,” said Andrew Cowell, Vice President, Mobility Infrastructure and IoT Power Business Division at Renesas. “Our multiphase PMIC employs Renesas’ industry-leading R5 modulation technology for blazingly fast transient response, which allows Google’s designers to dynamically scale power to improve overall system performance.”

Kai Yick, Google Hardware Manager said: “Renesas’ Multiphase PMICs provide our engineering team with the high performance and design flexibility required to build Coral. Our collaboration with Renesas ensures our Edge TPU achieve the highest power efficiency to perform calculations offline and locally.”