Microchips are used in practically all modern devices, including refrigerators, laptops, and phones. However, the process of creating them is complex behind the scenes. To make things easier, however, scientists claim to have discovered a method to harness the power of quantum computing.
A method of quantum machine learning, which combines the concepts of quantum computing with artificial intelligence (AI), has been created by Australian scientists and has the potential to revolutionize the production of microchips.
In a recent research that was released on June 23 in the journal Advanced Science, they presented their findings. In it, the researchers showed for the first time how quantum machine learning algorithms may greatly enhance the difficult task of simulating a chip’s internal electrical resistance, a crucial component that shapes its performance.
A hybrid strategy that combines quantum computing techniques with classical data is called quantum machine learning. Bits encoded as 0 or 1 are used to store data in classical computing. Qubits, which are used in quantum computers, may exist in numerous states at once because of concepts like superposition and entanglement. For example, two qubits can be 00, 01, 10, and 11 at the same time.
Complex mathematical connections may now be processed by quantum computing systems far more quickly than by traditional ones because to this; the more qubits a system has, the more parallel processing it can do.
Classical data is transformed into quantum states for quantum machine learning to use. This allows the quantum computer to find patterns in the data that would be difficult for classical systems to find. The results are then interpreted or used by a classical system.
Inside the process of producing chips
The manufacture of semiconductors is a multi-step, intricate process that demands meticulous accuracy at every stage. The smallest misalignment might lead to a chip malfunction.
The first step is to stack and shape hundreds of thin layers onto a silicon wafer, which is a thin, round piece of silicon that serves as the chip’s base.
Deposition creates thin layers of material on the wafer. Photoresist coating uses a light-sensitive substance to enable accurate patterning, which is the process of producing the small, complicated structures that characterize a chip’s circuitry.
Light is used in lithography to transfer such patterns to the surface of the wafer. Then, in order to create circuit architectures, etching eliminates certain regions of material. By inserting charged particles, ion implantation modifies each layer’s electrical characteristics. Lastly, the chip is packed, which entails connecting and encasing it in order to include it into a device.
This is where quantum computing ideas come into play. The researchers concentrated their investigation on simulating Ohmic contact resistance, which is a particularly challenging obstacle in chipmaking. This is a measure of how easily electricity flows between a chip’s metal and semiconductor layers; the lower the value, the quicker and more energy-efficient the performance may be.
Following the layering and patterning of the materials onto the wafer, this phase is crucial in defining the functionality of the final chip. However, precisely modeling it has proven to be challenging.
For this type of computation, engineers usually use traditional machine learning algorithms, which identify patterns in data to generate predictions. Machine learning may not be sufficient for semiconductor studies, which frequently provide small, noisy datasets with nonlinear patterns, even while this approach performs well with big, clean datasets. The team used quantum machine learning to solve this problem.
A novel type of algorithm
The group used data from 159 experimental samples of gallium nitride high-electron-mobility transistors (GaN HEMTs), which are fast and efficient semiconductors that are often employed in 5G devices and electronics.
In order to reduce the dataset to the most pertinent inputs, they first determined which manufacturing factors had the most effects on Ohmic contact resistance. Next, they created the Quantum Kernel-Aligned Regressor (QKAR), a novel machine learning architecture.
By transforming conventional data into quantum states, QKAR allows the quantum system to decipher intricate correlations within the data. These insights are then used by a classical algorithm to generate a prediction model that directs the fabrication of chips. They used five fresh samples that weren’t in the training set to test the model.
On these examples, the new model beat seven top classical models, including deep learning and gradient boosting approaches. QKAR produced a substantially better outcome than standard devices (0.338 ohm per millimeter) — albeit specifics were not provided in the study.
Crucially, though, it was made to work with hardware that is found in the actual world, which means that when quantum machines became more reliable, they may use it.
According to the study’s authors, these results show how [quantum machine learning] QML may be used to handle high-dimensional, small-sample regression tasks in semiconductor domains. They went on to say that the technique may soon be used to produce chips in the real world, especially when quantum technology develops further.
