Scientists in China have designed a tiny modular chip that is powered by light rather than electricity – and they want to use it to train and control a future artificial general intelligence (AGI) model.
The new chiplet, called “Taichi”, is one small piece of a wider mosaic made up of many individual chiplets (including Taichi modules) that together can form a complex and powerful computing system. If scaled up enough, it would be powerful enough to train and control AGI in the future, the scientists said in their paper published April 11 in the journal Science .
AGI is a hypothetical advanced form of artificial intelligence (AI) that would, in theory, be as smart as humans in terms of their cognitive reasoning abilities. AGI can be applied across many disciplines, while today’s AI systems can only be applied very narrowly.
Some experts believe such systems are many years away, with a key blocker being a bottleneck in computing power, while others believe we’ll be building an AGI agent as soon as 2027.
In recent years, scientists have begun to reach the limits of conventional electronics-based components, especially given the growth of AI and the vast amount of energy required to service these increasingly demanding systems.
Connected: Light-powered computer chip can train AI much faster than electricity-powered components
Graphics processing units (GPUs) have emerged as key components in AI training systems because they are better at performing parallel computations than central processing units (CPUs). But the required levels of energy consumption become unsustainable as systems get larger, the scientists say.
Light-based components may be one way to overcome the limitations of conventional electronics – including energy efficiency issues.
With an eye to the light for superhuman AI
Scientists previously outlined the design of a a new type of photonic microchip in February , which uses photons, or particles of light, instead of electrons, to control transistors — tiny electrical switches that turn on or off when a voltage is applied. Generally speaking, the more transistors a chip has, the more computing power it has and the more power it requires to operate. Light-based chips are much less power intensive and can perform calculations much faster than traditional chips because they can perform calculations in parallel.
Current photonic chip architectures for AI models consist of hundreds or thousands of training parameters or variables. This makes them powerful enough for basic tasks like pattern recognition, but large language models (LLMs) like ChatGPT are trained using billions or even trillions of parameters.
An AGI agent will likely require much more – as part of a wider network of AI architectures. Today, blueprints for building an AGI system do not exist.
In the new study, the scientists designed Taichi to work in the same way as other light-based chips, but could be scaled much better than competing designs, they said in their paper. That’s because it combines several advantages of existing photonic chips — including “optical diffraction and interference,” which are ways to manipulate light in the component.
To test the design, the researchers connected several Taichi chiplets and compared their architecture to other light-based chips in key areas.
Their architecture achieved a network scale of 13.96 million artificial neurons—compared to 1.47 million in the next largest competing design—with a power efficiency metric of 160.82 trillion operations per watt (TOPS/W). The next best result they highlighted in their paper came from research published in 2022 , in which the photonic chip achieved 2.9 TOPS/W. Many conventional neural processing units (NPUs) and other chips achieve well below 10 TOPS/W .
The researchers also claim that their Taichi-based architecture is twice as powerful as other photonic systems, but they don’t cite them directly. Meanwhile, in tests, they used the Taichi distributed network to perform tasks including image categorization and classification, as well as image content generation, as a proof of concept rather than a performance comparison.
“Taichi points to the great potential of photonic computing on a chip to handle a variety of complex tasks with large network models, enabling real-world applications of optical computing,” the scientists said. “We expect Taichi to accelerate the development of more powerful optical solutions as critical support for the core model and a new era of AGI.”
