Core Viewpoint - The company is actively developing AI technology in collaboration with various institutions, focusing on NPU (Neural Processing Unit) IP cores for edge applications, with several products already in design and development stages [1] Group 1: NPU Development - The company is collaborating with the Hong Kong Applied Science and Technology Research Institute to develop AI technology for edge applications, resulting in the CNN20, CNN100, and CNN200 series NPU IP cores [1] - The CNN20 and CNN100 have completed design and are available for external licensing, while the CNN200 is still under development [1] - The CNN20 and CNN100 have a single-core computing power of 1 Tops@INT8, suitable for low-power AI MCU chips, while the CNN200 can reach 10 Tops@INT8, applicable for various edge computing AI SoC chips [1] Group 2: AI Applications - The NPU IP cores can be widely applied in numerous AI scenarios, including applications like robotic dogs [1] - The company is also collaborating with Longqing Technology to develop the CNN300 NPU IP core for AI PC applications, which combines scalar and vector processing units [1] - The CNN300 is designed to achieve a single-core performance of 8 Tops, with multi-core consistency interface MLS, allowing for higher computing power through multi-core stacking, potentially reaching 32 Tops with four cores [1] Group 3: RISC-V GPGPU Development - The company is working with the Tsinghua International Innovation Center in Shanghai on GPGPU (General-Purpose Graphics Processing Unit) core development based on the open-source RISC-V instruction architecture [1] - The first version of the RTL code design for the RISC-V GPGPU has been completed and publicly released as of July 2024 [1] - The current phase of RISC-V GPGPU development has concluded, with future plans dependent on the international development of open-source RISC-V GPGPU technology [1]

Core Insights - Company is collaborating with Hong Kong Applied Science and Technology Research Institute in the NPU field to develop AI technology for edge applications, resulting in the CNN20, CNN100, and CNN200 series NPU IP cores [1] - CNN20 and CNN100 have completed design and are available for external licensing, while CNN200 is still under development [1] - CNN20 and CNN100 have a single-core computing power of 1 Tops@INT8, suitable for low-power AI MCU chips, while CNN200 can reach 10 Tops@INT8, applicable for various edge computing AI SoC chips [1] - The company is also working with Longqing Technology to develop the CNN300 NPU IP core for AI PC applications, which combines scalar and vector processing units, achieving a single-core performance of 8 Tops [1] - The CNN300 features a multi-core consistency interface (MLS) that supports expansion for higher computing power, allowing for a potential 32 Tops performance with a four-core stack [1] - Collaboration with Tsinghua International Innovation Center in Shanghai is focused on developing GPGPU cores based on the open-source RISC-V instruction architecture, with the first version of RTL code design completed and publicly released by July 2024 [1] - The company is concentrating on the R&D and design of automotive electronic chips, server and cloud application security chips, quantum security chips, and AI MCU chips, while the current phase of RISC-V GPGPU development has concluded [1] - Future R&D plans for RISC-V GPGPU will depend on the further development of open-source RISC-V GPGPU technology internationally [1]

Core Insights - Company is collaborating with Hong Kong Applied Science and Technology Research Institute in the NPU field to develop AI technology for edge applications, resulting in the CNN20, CNN100, and CNN200 series NPU IP cores [1] - CNN20 and CNN100 have completed design and are available for external licensing, while CNN200 is still under development [1] - CNN20 and CNN100 have a single-core computing power of 1 Tops@INT8, suitable for low-power AI MCU chips, while CNN200 can reach 10 Tops@INT8, applicable for various edge computing AI SoC chips [1] - The company is also working with Longqing Technology to develop the CNN300 NPU IP core for AI PC applications, which combines scalar and vector processing units, achieving a single-core performance of 8 Tops [1] - The CNN300 features a multi-core consistency interface (MLS) that supports the expansion of multiple CNN300 IP cores for higher computing power, with a potential 32 Tops performance using a four-core stack [1] - Collaboration with Tsinghua International Innovation Center in Shanghai is focused on developing GPGPU cores based on the open-source RISC-V instruction architecture, with the first version of RTL code design completed and publicly released by July 2024 [1] - The company is concentrating on the R&D and design of automotive electronic chips, server and cloud application security chips, quantum security chips, and AI MCU chips, while the current phase of RISC-V GPGPU development has concluded [1] - Future R&D plans for RISC-V GPGPU will depend on the further development of open-source RISC-V GPGPU technology internationally [1]

Core Insights - Guoxin Technology (688262.SH) has introduced the CNN200, which utilizes a GCU+NN network architecture, achieving a maximum single-core computing power of 10 TOPS@INT8, suitable for various edge computing AI SoC chips and applications including robotic dogs [1] Group 1: Product Features - The CNN200 features a pulsed array computing unit that optimizes dynamic power consumption and memory area, effectively reducing energy consumption and latency, achieving industry-leading energy efficiency [1] - It integrates on-chip caching and inter-layer data sharing technology, significantly reducing DDR access [1] - The hardware acceleration unit supports over 90 types of neural network operators and has a quick expansion interface design to adapt to the development of neural network models [1] Group 2: Compatibility and Ecosystem - The CNN200 supports post-training quantization (PTQ) with various quantization strategies, including symmetric, asymmetric, layer-wise, and channel-wise, compatible with mainstream neural network structures like CNN and RNN, and supports INT8 and FP16 data precision [1] - It is compatible with major deep learning frameworks such as PyTorch, TensorFlow, ONNX, and PaddlePaddle, demonstrating broad ecological adaptability [1] - The accompanying NPU toolchain includes tools for model format conversion, preprocessing, quantization, compilation, and simulation, providing software ecosystem support for NPU inference implementation and application deployment [1] Group 3: Future Developments - The company is developing the CNN300, aimed at AIPC applications, which is expected to support data types such as INT8, FP8, and FP16, and will cater to both traditional CNN and RNN applications as well as the latest popular LLM (large language model) applications [1] - The CNN300 will facilitate the offloading of commonly used large models like Deepseek, Qwen, and LLaMa, meeting the needs of conventional applications such as speech, image, and video recognition, as well as high-quality speech and video display for AIPC applications, generative AI, multimodal interaction, and knowledge management [1]