VeriSilicon’s Ultra-Low Energy NPU Provides Over 40 TOPS for On-Device LLM Inference in Mobile Applications

The energy-efficient architecture scales across AI-enabled devices, including AI phones and AI PCs

Shanghai, China, June 9, 2025--VeriSilicon (688521.SH) today announced that its ultra-low energy and high-performance Neural Network Processing Unit (NPU) IP now supports on-device inference of large language models (LLMs) with AI computing performance scaling beyond 40 TOPS. This energy-efficient NPU architecture is specifically designed to meet the increasing demand for generative AI capabilities on mobile platforms. It not only delivers powerful computing performance for AI PCs and other end devices, but is also optimized to meet the increasingly stringent energy efficiency challenges of AI phones and other mobile platforms.

Built on a highly configurable and scalable architecture, VeriSilicon’s ultra-low energy NPU IP supports mixed-precision computation, advanced sparsity optimization, and parallel processing. Its design incorporates efficient memory management and sparsity-aware acceleration, which reduce computational overhead and latency, ensuring smooth and responsive AI processing. It supports hundreds of AI algorithms including AI-NR and AI-SR, and leading AI models such as Stable Diffusion and LLaMA-7B. Moreover, it can be seamlessly integrated with VeriSilicon’s other processing IPs to enable heterogeneous computing, empowering SoC designers to develop comprehensive AI solutions that meet diverse application needs.

VeriSilicon’s ultra-low energy NPU IP also supports popular AI frameworks such as TensorFlow Lite, ONNX, and PyTorch, thereby accelerating deployment and simplifying integration for customers across various AI use cases.

“Mobile devices, such as smartphones, are evolving into personal AI servers. With the rapid advancement of AIGC and multi-modal LLM technologies, the demand for AI computing is growing exponentially and becoming a key differentiator in mobile products,” said Weijin Dai, Chief Strategy Officer, Executive Vice President, and General Manager of the IP Division at VeriSilicon. “One of the most critical challenges in supporting such high AI computing workloads is energy consumption control. VeriSilicon has been continuously investing in ultra-low energy NPU development for AI phones and AI PCs. Through close collaboration with leading SoC partners, we are excited to see that our technology has been realized in silicon for next-generation AI phones and AI PCs.”

About VeriSilicon

VeriSilicon Microelectronics (Shanghai) Co., Ltd. (VeriSilicon, 688521.SH) is committed to providing customers with platform-based, all-around, one-stop custom silicon services and semiconductor IP licensing services leveraging its in-house semiconductor IP.

VeriSilicon possesses six categories of in-house processing IPs, namely Graphics Processing Unit (GPU) IP, Neural Network Processing Unit (NPU) IP, Video Processing Unit (VPU) IP, Digital Signal Processing (DSP) IP, Image Signal Processing (ISP) IP, and Display Processing IP, as well as more than 1,600 analog and mixed-signal IPs and RF IPs.

Leveraging its own IPs, VeriSilicon has developed a wealth of software and hardware custom chip design platforms targeting Artificial Intelligence (AI) applications, covering always-on ultralight spatial computing devices such as smartwatches and AR/VR glasses, high-efficiency edge computing devices such as AI PCs, AI phones, smart cars, and robots, as well as high-performance cloud computing devices like data centers and servers.

In response to the trend of System-on-Chip (SoC) evolving towards System-in-Package (SiP) driven by the demand for large computing power, VeriSilicon put forward the concepts of "IP as a Chiplet”, "Chiplet as a Platform", and "Platform as an Ecosystem”. The company keeps advancing the R&D and industrialization of its Chiplet technologies and projects from the perspective of interface IP, Chiplet architecture, advanced packaging technology, and others for AI-Generated Content (AIGC) and autonomous driving solutions.

Under its unique “Silicon Platform as a Service” (SiPaaS) business model, VeriSilicon serves a broad range of market segments, including consumer electronics, automotive electronics, computer and peripheral, industry, data processing, Internet of Things (IoT), among others. Its main customers include fabless, IDM, system vendors (OEM/ODM), large internet companies, and cloud service providers.

Founded in 2001 and headquartered in Shanghai, China, VeriSilicon has 8 design and R&D centers, along with 11 sales and customer service offices worldwide. VeriSilicon currently has more than 2,000 employees.