AI chips can be reconfigured and upgraded on demand


　　Future mobile phones can be built like Lego blocks

　　Science and Technology Daily, Beijing, June 13 (Reporter Zhang Mengran) Imagine a more sustainable future: your mobile phone, smartwatch and other wearable devices do not have to be put on hold or discarded for replacement.

Instead, they can be upgraded with the latest sensors and processors, which fit on internal chips and integrate like Lego bricks.

MIT engineers have used a Lego-like design to create a stackable, reconfigurable artificial intelligence (AI) chip.

The chip building blocks keep devices up-to-date while reducing waste of electronics.

The research was published in Nature Electronics.

　　The new design uses light instead of physical wires to transmit information through the chip.

Therefore, you can add as many computational layers and light, pressure and even smell sensors as you want.

The researchers call it a Lego-like reconfigurable AI chip because of its infinite scalability depending on the combination of layers.

　　In the new chip design, the researchers paired image sensors with arrays of artificial synapses, and trained each synapse array to recognize certain letters -- in this case, M, I, and T. An optical system is fabricated between the arrays to enable communication between layers without the need for physical connections, so chips can be freely stacked and added in any desired way.

　　Photodetectors can form an image sensor that receives data and transmits the data to the LEDs in the next layer.

When a signal, such as an image of a letter, reaches the image sensor, it stimulates another layer of photodetectors as well as an array of artificial synapses, which classifies the signal based on the pattern and intensity of incoming LED light.

　　The team built a chip based on this, stacking three image-recognition "blocks", each including an image sensor, optical communication layer and an array of artificial synapses, which are used to identify the three letters M, I or T. one for classification.

They then shined pixelated images of random letters onto the chip and measured the current produced by each neural network array in response.

The higher the current, the more likely it is that the image is indeed a specific array of recognized letters.

　　This design demonstrates stackability, replaceability, and the ability to insert new functionality into the chip.

The researchers envision making a general-purpose chip platform, where each layer could be sold individually like a video game; or making different types of neural networks, such as image or speech recognition, letting customers choose what they want and adding them like Lego bricks into existing chips.

　　[Editor-in-chief's circle]

　　This innovation is indeed quite attractive. There is no need to purchase a new phone, and the mobile phone chip can be directly upgraded.

Each new generation of electronics advertises itself as "faster and stronger," and one of the secrets is more advanced processors.

Researchers have created a reconfigurable chip whose key structures are image sensors, an array of artificial synapses, and an optical communication system.

Synaptic arrays can be trained to have multiple functions, responding to different images.

This kind of chip is more flexible and can be stacked layer by layer; it can also strengthen the ability of a certain aspect as needed to meet the needs of customers in different scenarios.

Just don't know when they will reach commercial maturity.