AIoT product proliferation creates an increasing demand for a flexible, performant processing core that offers short time-to-market. Demands around privacy and user-experience are pushing tasks from Cloud to Edge (on-device) processing and challenging the capabilities of traditional processors.
With xcore.ai, real-time inferencing, decisioning at the edge, signal processing, control and communications are wrapped up in a single chip. Which means Product Designers no longer need to rely on a costly applications processor or a microcontroller with additional components.
xcore.ai combines AI acceleration, powerful DSP, connectivity and general-purpose compute in a low eBOM solution - ideal for developers working on smart products that enhance our everyday living.
The device has two “tiles”. Each tile is a self-contained processor with 512 kByte single cycle SRAM. The tile has a scalar unit (up to 1600MIPS), a vector unit (up to 25,600 MMACS), and a floating point unit (up to 800 MFLOPS); 1 Mbyte tightly coupled SRAM, 3200 MIPS, 1600 MFLOPS, and 51,200 MMACCS across the tiles. The device has three integrated PHYs: a high-speed USB, a MIPI D-PHY receiver, and LPDDR1.
Supports 1.8V and 3.3V on each bank of I/O, allowing the exact mix of peripheral interfaces and connections for your system. With USB, MIPI RX and a software driven I/O structure supporting I2C, I2S, S/PDIF, SPI, UART, PWM, and many more, the xcore.ai offers a single chip solution (replacing the need for multi-components).
Convolutional (CNN) and Dense (DNN) Neural Network acceleration supports 32-bit 16-bit, 8-bit and 1-bit vector operations. Instructions loading a vector from memory, multiplication and accumulation occur in a single clock cycle (achieving 51.2 GMACC/s peak, and 30GMACC/s sustained 8-bit vector arithmetic).
Industry standard tools enable accelerated applications. xcore.ai is fully programmable in C and C++, with optimised libraries for rapid prototyping and development. FreeRTOS support and deep learning framework TensorFlow-Lite ensures developers will enjoy an easy familiarity with the programming model.
With up to 3200MIPS of compute, the fast and predictable xcore.ai can handle the most challenging edge-AI operations. With up to 8 FIR taps (32’bit) per clock cycle and 1 million 512 FFTs per second, it’s ideal for voice and sensor processing applications.