2024-03-28 20:41:08
A chip that uses photons — rather than electrons — to perform complex calculations could overcome the limitations of classical silicon chip architecture and significantly speed up the processing speed of computers — while reducing their energy consumption, according to a new study by researchers published in žurnale „Nature Photonics“.
Silicon chips contain transistors, tiny electrical switches that turn on or off when a voltage is applied. The more transistors a chip has, the more computing power it has – but the more power it needs.
Throughout the history of computing, chips have been built according to Moore’s Law, which doubles the number of transistors every two years (but without increasing manufacturing costs or power consumption). But silicon chips have physical limitations — including the maximum speed at which transistors can operate, the heat they emit, and the smallest chip size scientists can make.
This means that packing billions of transistors into ever-smaller silicon chips may be impossible.
However, using photons has many advantages over electrons. First, they move faster than electrons, which cannot reach the speed of light. Although electrons can move at almost this speed, for such systems would require an enormous – and unfeasible – amount of energy. Therefore, the use of light would require much less energy. Also, photons have no mass and do not emit heat like electrically charged electrons.
In developing the chip, the researchers aimed to create a light-based platform that could perform calculations called vector matrix multiplication. It is one of the basic mathematical operations used to train neural networks, machine learning models designed to mimic the architecture of the human brain. This is how AI tools like ChatGPT and Google Gemini are trained.
Instead of using a single-height silicon wafer for the semiconductor, as is done in conventional silicon chips, the researchers made the silicon thinner – but only in certain areas.
“These changes in height – without the addition of any other materials – provide control over the propagation of light in the chip, as the changes in height can be distributed so that the light scatters in specific patterns, allowing the chip to perform mathematical calculations at the speed of light,” – says one of the lead authorsNader Engheta, professor of physics at the University of Pennsylvania (USA).
The researchers say that their project can be adapted to already existing production methods, it does not need to be adapted. That’s because the researchers used the same techniques used to make conventional chips to make the photonic chip.
The researchers also say that the project’s schematics can be applied to the development of graphics processing units (GPUs), which have seen a significant increase in demand in recent years. This is because these components are central to training large language models (LLMs) such as Google Gemini or OpenAI ChatGPT.
“They can adapt the Silicon Photonics platform as an add-on,” says Firooz Aflatouni, a professor of electrical engineering at the University of Pennsylvania in the US and one of the authors of the study. – And then it would be possible to speed it up [dirbtinio intelekto] training and classification”.
Let’s talk about „Live Science“.
2024-03-28 20:41:08