Home | Markets & Applications | AI

Artificial Intelligence

The development of the third wave of artificial intelligence is driven by the exploration of “deep learning (DL)” technology, which breaks the deadlock of the difficulty of deep network training. In addition, coupled with the improvement of GPU computing power, after 2011, in the fields of speech and image recognition respectively breakthrough progress has been achieved, paving the way for the successful development of deep learning. Furthermore, in 2020, OpenAI created the third-generation language model GPT-n, which generates text that is like human beings and has made rapid progress, thereby stimulating a new wave of artificial intelligence development. The rapid development of AI has profoundly changed our lifestyle and is shaping the new pattern of the global economy. AI is not only an innovation in science and technology but will also generate huge business opportunities at all levels. According to a Precedence Research research report, it is expected that by 2030, the global AI market size will reaching US$1,597.1 billion.

With the rapid development of new technologies such as cloud computing and 5G, AI computing power continues to grow and gradually expands from the data center to the end point, forming a three-level computing power architecture of cloud-edge-end. In addition, due to the diverse processing requirements of sound/image/data, different computing tasks need to be performed by dedicated processors that meet the needs to enhance computing performance and improve energy efficiency. CPU + heterogeneous computing (GPU, DPU, VPU, …), computing units are interconnected through high-traffic, low-latency channels, which have the advantages of versatility and low energy consumption. RISC-V provides an extensible instruction set, which provides more flexibility than x86 and Arm ISA, and can have advantages in performance, power consumption, and scalability. Andes has DSP/SIMD and vector RISC-V processors for different AI applications and many customers are in volume production. Together with the AnDLA neural-network accelerator, Andes solutions help companies meet the diverse AI/ML computing requirements from edge to the cloud. Below is a brief description of the Andes AI/ML solutions:

Andes RISC-V DSP/SIMD Processor:

AndesCore™ D25F: 32-bit/5-stage pipeline/in order/single issue, it can deliver high per-MHz performance with small gate count, suitable for networking and communications, video and image processing, sensor device/sensor hub, advanced motor control, etc.

AndesCore™ D45: 32-bit/8-stage pipeline/in order/dual issue, it is equipped with comprehensive SIMD/DSP instructions that can boost the performance of voice, audio, image, and signal processing, suitable for networking and communications, smart wireless switch/router, advanced industry controller, etc.

Andes RISC-V Vector Processor:

AndesCore™ NX27V:  64-bit/5-stage pipeline/in order/single issue/vector processing unit, it brings enhanced performance in memory subsystem with higher memory bandwidth and memory latency reduction by supporting multiple outstanding data access, suitable for networking and scientific computing, multimedia processing, machine learning, Deep Learning, AR/VR, etc.

AndesCore™ AX45MPV: 64-bit/8-stage pipeline/in order/dual issue/vector processing unit, it features MMU for Linux based applications, level-1 instruction/data caches and local memories for low-latency accesses, suitable for computer vision, scientific computing, digital signal processing, machine/deep learning acceleration, etc.

AndesAIRE™ (Andes AI Runs Everywhere) are ultimately efficient solutions designed for edge and end-point inference, a brief introduction follows:

AndesAIRE™ AnDLA™ I350: Deep Learning Accelerator (DLA), it supports popular deep learning frameworks, such as TensorFlow Lite, PyTorch, and ONNX, and accelerated engine for GEMM, MAC accumulator, element-wise, pooling, etc., suitable for AIoT device, smart camera, smart home appliance, robotic, etc.

AndesAIRE™ NN SDK: A collection of software tools, libraries, runtimes, and sample programs for end-to-end development and deployment of AndesAIRE™ AnDLA™ and AndesCore™ RISC-V CPU platforms.

AndesAIRE™ NN library:  A powerful and efficient neural network compute library that is designed to accelerate neural network tasks on AndesCore™ RISC-V CPU platforms.