Michael Tomlinson, an Electrical and Computer Engineering graduate, and Joe Li, an undergraduate of Johns Hopkins University, have created neural network chips, a nueromorphic accelerators, that works related to human brain, using prompts in ChatGPT4.
The neural network chips have potentials than you ever think of. The chip is advance enough to power energy-efficient, real-time machine intelligence for next-generation embodied systems like autonomous vehicles (popular known as driverless car) and human-mimicked robots (humanoid).
Michael and Joe are students from the Johns Hopkins University and they’re also members of the Andreou Lab, a research lab in the Johns Hopkins University, Maryland. They used natural language prompts and OpenAI’s ChatGPT4 to produce detailed instructions to incept the neural network chip.
Through step-by-steps entry of prompts into GPT4 to form a network, they generated a full chip design that could be fabricated, making it “the first AI chip that is designed by a machine using natural language processing.”
How does the Chip works
The chip will be produced or fabricated into a tiny silicon brain that have two layers of interconnected neurons. You can increase or reduce the power of the chip connections using an 8-bit addressable weight system, allowing the chip to configure learned weights that determine the chip’s functionality and behavior.
You can also reconfigure and program the chip using a user-friendly interface called “Standard Peripheral Interface (SPI)” sub-system. This system is like a remote control, it’s also built by ChatGPT using natural language prompts.
Final words
This nueromorphic chip was designed without a single complex coding as a proof of concept, said Tomlinson. However, before manufacturing the chip, the Tom and Joe tested it through extensive software simulations to ensure that the end product would work as expected.
After manufacturing, the chip was reviewed electronically by the Skywater “foundry,” a chip fabrication service where it is currently being “printed” using a relatively low-cost 130-nanometer manufacturing CMOS process.
Overall, this discovery has unveils that artificial intelligence (AI) can be used to create advanced AI hardware systems that would help boost AI technology ecosystems and usage.