Tsinghua University Develops 'Taiji-II' Optical Training Chip for Efficient Training of Large-Scale Neural Networks in Optical Computing Systems

AIbase基地

Published in AI News · 4 minute read · Jul 19, 2025

Recently, Tsinghua University has achieved a significant breakthrough in the field of optical computing, successfully developing the "Taiji-II" optical training chip. This advancement enables efficient and precise training of large-scale neural networks in optical computing systems. This innovative achievement was jointly completed by the research team led by Professor Fang Lu from the Department of Electronic Engineering and Academician Dai Qionghai from the Department of Automation at Tsinghua University. They pioneered a full-forward intelligent optical computing training architecture, opening new avenues for the development of optical computing technology.

GPU Chip (6)

Image source: Picture generated by AI, authorized service provider Midjourney

In recent years, intelligent optical computing has attracted significant attention in the field of computing power development due to its high computational power and low power consumption. The previous general-purpose intelligent optical computing chip "Taiji" has moved optical computing from principle verification to large-scale experimental applications, achieving a system-level energy efficiency of 160TOPS/W. However, existing optical neural network training still heavily relies on GPUs for offline modeling and requires precise alignment of the physical system, which limits the further development of optical computing technology.

The successful development of the "Taiji-II" optical training chip overcomes these limitations. According to Tsinghua University, this new chip can train various optical systems and demonstrate excellent performance across multiple tasks. It no longer relies on GPUs for offline modeling, simplifying the training process and providing an efficient and precise new method for large-scale neural network training in optical computing systems.

QQ Screenshot 20240808135353.png

This breakthrough is of great significance. It not only showcases China's research strength and innovative capabilities in cutting-edge technologies but also has the potential to promote the widespread application of optical computing technology in areas such as artificial intelligence and big data processing. With the advent of the "Taiji-II" optical training chip, optical computing technology will see significant improvements in training capabilities and application scope, paving the way for the development of more efficient and intelligent computing systems in the future.

Paper link: https://www.nature.com/articles/s41586-024-07687-4

Solving the Training Dilemma of Llama3! Doubao's Large Model Teams Up with Hong Kong University to Launch a New Checkpoint System to Optimize Training Efficiency

In the digital world, the growth of artificial intelligence relies on checkpoints, which can restore the training status to the last safe state during training large language models (LLMs) if issues like sudden power outages or hardware failures occur. However, traditional checkpoint systems are inefficient when dealing with large models. To address this, ByteDance and the research team from Hong Kong University have proposed ByteCheckpoint, an innovative checkpoint system that improves data and metadata handling.

US Defense Startups Raise $1.5 Billion to Produce Tens of Thousands of AI Weapons

Anduril, founded by Palmer Luckey, has recently secured $1.5 billion in investment and plans to establish a smart factory for the production of autonomous weapons and advanced battlefield technologies, aiming for an annual output of tens of thousands of units. The company aims to create a highly efficient manufacturing platform similar to Tesla's, optimizing weapon system design and operation through AI technology to enhance combat efficiency and reshape military strategy. Anduril's rise has garnered significant attention in the defense technology sector, signaling a revolutionary change in the integration of technology and military.

Tencent Hunyuan Leads in Multimodal AI: Comprehensive Advantage Over GPT-4/Claude-3.5

Domestic large models have shown strong capabilities in the multimodal field of artificial intelligence. Tencent's Hunyuan-vision and Shanghai AI Lab's InternVL2-40B are leading their respective closed-source and open-source fields, surpassing internationally renowned models. Hunyuan-vision, as the multimodal version of Tencent's Hunyuan large model, is favored by developers for its API calls and is offered for free in Tencent's Yuanbao APP, showcasing practicality and ease of use. Yuanbao APP ranks first in China for multimodal capability assessments through a series of tests.

Baidu Announces 'Wenxin Kuai Ma' as the First AI Architect in the Internet

'Wenxin Kuai Ma', an AI product from Baidu, has elevated to the position of AI Architect just a year and a half after joining Baidu in April 2023, making it the first AI Architect in the internet industry. As an intelligent code assistant based on the Wenxin large model, Wenxin Kuai Ma significantly enhances development efficiency, with a code adoption rate reaching 44%, generating 33% of the company's code daily. Its three core skills include a deep understanding of business scenarios, providing code quality assurance, and intelligent debugging, which notably improves product quality and developers' problem-solving capabilities. Wenxin Kuai Ma has successfully integrated with several companies.

Alibaba Cloud Launches First AI Model Application for Domain Names, Allowing One-Click Generation of Creative Domains

Alibaba Cloud announced the AI transformation of its domain product services at the Wanwang Renewal Launch Conference and introduced the first domain AI application based on the Tongyi large model. Users can now generate creative domain names with one click by entering brand information and industry. The AI will perform semantic analysis, evaluate the meaning of the domain, and filter available options.

Flip-Flopping? UK New Government Shelves $170 Million AI Project Funding

The newly elected Labour government in the UK recently promised to prioritize artificial intelligence within its policy framework. However, while trying to balance the national budget, it unexpectedly shelved $170 million originally intended for AI and technology projects. This move stems from a fiscal gap left by the previous government, with new Chancellor Rachel Reeves accusing her predecessors of 'covering up' a $27.9 billion public spending gap. In the context of budget cuts, many AI projects have been impacted, especially the $1 billion petascale supercomputer project based in Edinburgh.