Hybrid device merges DSP and MCU architectures

0
81




Embedded Staff


January 08, 2019




embedded.com.staff-January 08, 2019

CEVA has announced its CEVA-BX hybrid processor and WhisPro speech recognition technology. CEVA-BX combines both digital-signal processor (DSP) and microcontroller (MCU) features in a single device designed to fill a growing need for both signal-processing and computing capabilities in a range of application areas including IoT, consumer, automotive, and industrial. According to the company, the new family addresses a gap in the signal-processing performance of MCUs and the flexibility of DSPs that has limited designers ability to easily address emerging requirements for cellular IoT, sensor fusion, neural-network inference, and more. 

Although DSPs, MCUs, and specialized processors remain the best choice for most well-defined workloads, emerging applications present mixed workloads. As a result, developers find they either need to combine MCUs and DSPs in more complex designs or accept compromises in signal-processing performance or control flexibility. The CEVA-BX family is designed to support mixed workload applications through an architecture that combines parallel-processing capability with advanced microprocessor features (Figure 1). 

click for larger image

Figure 1. CEVA-BX block diagram. (Source: CEVA)

Built around an variable-length pipeline (11 stages max), the architecture combines two scalar procssing units (SPUs) with native support for complex math, FFT, division acceleration, and optional support for double-, single-, and half-precision floating-point units. In addition, the instruction set supports single-instruction multiple-data (SIMD) operations required for vector processing in neural-network inference and other algorithms. To support MCU-type control operations, the architecture combines an instruction-set architecture with full C-type support with a large general-purpose register file and buffers for branching and loops designed to reduce code size and speed performance. The fully cached memory subsystem supports 4GB program and data space, providing full control of memory pages and dedicated master ports for each. For even more complex applications, developers can use the CEVA-Connect’s automatic queue and buffer management mechanisms to integrate co-processors and create a cluster of CEVA-BX cores (Figure 2). 


Figure 2. CEVA-Connect data transfers. (Source: CEVA)

The CEVA-BX is initially offered in two configurations – the CEVA-BX1 with single 32X32-bit MAC and quad 16X16-bit MACs and the CEVA-BX2 with quad 32X32-bit MACs and octal 16X16-bit MACs, that are also capable of supporting 16×8-bit and 8×8-bit MAC operations. The CEVA-BX2 addresses intensive workloads such as 5G PHY control, multi-microphone beamforming and neural networks for speech recognition, with up to 16 GMACs per second. The CEVA-BX1 serves low to mid-range DSP workloads, such as cellular IoT, protocol stacks, and always-on sensor fusion, with up to 8 GMACs per second. Security is addressed using dedicated trusted execution modes to comply with the stringent safety standards. The CEVA-BX family is accompanied by a comprehensive software development tool chain, including an advanced LLVM compiler, Eclipse based debugger, DSP and neural network compute libraries, neural network frameworks support such as Android NN API, ARM NN, and Tensorflow Lite, and choice of industry leading Real Time Operating Systems (RTOS). For more information, visit the CEVA-BX product page.

WhisPro Speech Recognition

Separately, CEVA announced its WhisPro technology is designed to speed implementation of voice-activated smart products that connect to cloud-based voice assistant services, such as Amazon Alexa, Google Assistant, Baidu DuerOS, and others. Designed for always-on listening edge devices, the technology is built around a scalable recurrent neural network (RNN) model that is capable of handling a single trigger phrase, as well as simultaneous multi trigger phrases, for supporting multiple AI assistants. Designed with built-in noise immunity, WhisPro achieves a speaker-independent wake-phrase recognition rate over 95% while minimizing power consumption and processing requirements. For even greater noise immunity, developers can combine WhisPro with CEVA’s ClearVox noise-reduction front-end technology. Working in tandem, WhisPro and ClearVox provide a noise-tolerant voice-activation software solution designed to run on CEVA CEVA-TeakLite-4, CEVA-X2, and CEVA-BX devices. For more information, visit the CEVA WhisPro product page. 

 

 




Hybrid device merges DSP and MCU architectures