Quantum Computer 'Pruning' Inspired by Pretty Little Nematode...

Seoul National University Research Team Develops Technology to Enhance Reliability of Photonic Circuits

Domestic researchers have succeeded in significantly improving the reliability of photonic circuits used in quantum computers and artificial intelligence (AI) deep learning by employing a pruning method that selectively retains only essential components.

The National Research Foundation of Korea announced on the 26th that the research team led by Professors Namkyu Park and Seongyu Yoo at Seoul National University developed a quantum circuit pruning technique that dramatically enhances the reliability of universal quantum computers and photonic machine learning.

Image source provided by the National Research Foundation of Korea.

In future computing, the role of light is gaining attention as an innovative technology that provides massive processing power. Light enables ultra-fast and low-loss computations. For implementing universal computing not limited to specific problems, the use of photonic integrated circuits capable of real-time control and programming of light states is essential. However, as the circuit scale increases, thermal noise from components significantly reduces computational reliability, posing a major challenge in scaling the number of quantum qubits and deep learning neurons to commercially viable levels.

The research team overcame challenges in the photonic circuit field through a multidisciplinary approach. Inspired by the nematode Caenorhabditis elegans, which performs remarkable functions with very few neurons, and aviation networks where hub components drive system operation, they attempted a more efficient photonic hardware implementation than before. During the hardware analysis of photonic circuits used in quantum computing and AI, they confirmed that the Pareto principle (80/20 rule) from network science also applies to photonic circuits. This means that hub components exist separately, and by removing less important components, high-reliability and low-power circuits can be realized.

To engineer the hidden inequalities found in nature, the research team introduced the concept of pruning, commonly used in deep learning software, into photonic hardware design for the first time, developing a 'quantum circuit pruning' technique. They demonstrated that this enables the implementation of highly reliable quantum computing and deep learning accelerators. Pruning is a lightweight technique that retains important parameters during model training while removing less important ones.

Professor Park explained, “This research can be described as minimalism in quantum circuits that leaves only the necessary components,” adding, “It is very encouraging that the efficiency of pruning improves further in large-scale quantum computing and deep learning accelerators.”

The results of this study were published on the 3rd in the international journal Nature Communications.

This content was produced with the assistance of AI translation services.