Tech Talk: How to cope with the increasing need of computing power?

Editor's note: Dive into the vision of Chu Junhao, a member of the Chinese Academy of Sciences and semiconductor specialist, where AI meets semiconductor evolution. This interview unravels the intertwined future of technology. With Moore's Law decelerating, Chu proposes a reevaluation of our approach and advocates for a synergy of human intellect and cutting-edge hardware innovations. Join us in delving into the AI-semiconductor partnership and the critical search for innovative solutions in an era of increasing digital needs.

With information technology advancement in AI becoming integrated across various industries, the demand for computing power is set to increase because of AI's inherent reliance on extensive calculations for its operations, said Chu Junhao, a member of the Chinese Academy of Sciences and semiconductor specialist, during a recent interview with CGTN.

AI is expected to increasingly merge with sectors such as the medical industry, materials science, the information industry, and energy and health sectors in the future. This integration is poised to generate numerous new industrial opportunities. However, Chu emphasized that such advancements are contingent upon the availability of substantial computing power.

To address the growing demand, Chu, who is also a popular knowledge content creator on Bilibili, stresses the importance of prioritizing the enhancement and expansion of computational infrastructure.

"It's essential to strategically plan the distribution of computing power, both regionally and in its internal structure," he said. "Furthermore, boosting computing power through enhancements in semiconductor hardware technology is critical."

Such improvements necessitate innovation in hardware. For instance, advancements in compute-in-memory (CIM) technology could enable on-chip processing. This means that only the processed outcomes, rather than large quantities of unprocessed data, need to be transmitted, greatly reducing the need for storage.

"Within a large dataset, the goal is to isolate a single unit that holds a relatively minor amount of data because it has been pre-processed," Chu explained. "By organizing data into units or undergoing data unitization, we can proceed to the subsequent processing stage. This method also helps save on computing power."

As the industry explores technological advancements and strategies to meet increasing computing demands, it's crucial to acknowledge the broader context of semiconductor development, as illuminated by Moore's Law. This law predicts that the processing power of computers will double approximately every two years, a trend that has driven the exponential growth of computing capabilities for decades.

However, with semiconductor components shrinking to the brink of atomic scales, the challenge of sustaining this growth rate becomes apparent. This critical point in the industry, while not signaling an end to progress, highlights the imperative for new materials and innovative computing models, according to Chu.

The deceleration of Moore's Law reflects a pivotal shift in strategies for augmenting computing power and acts as a catalyst for exploring innovative approaches that will shape the future of AI, semiconductors, and the intersection of the two.

Chu highlighted that spintronic devices, along with quantum chips and photonic computing, are poised to play a pivotal role in the future of technology. "We are currently engaged in unraveling the scientific principles and practical applications of these technologies. Once fully implemented, our focus will shift towards their real-world applications, allowing us to pool our efforts in these areas."

While boosting computing power requires advances across several fronts like CIM technologies, as well as the development and application of quantum chips and photonic computing, Chu also highlighted the importance of incorporating human-like intelligence into hardware as a critical step for future breakthroughs in AI and computing capabilities.

He cited an example where a chess-playing robot's gameplay can consume a vast amount of electricity, potentially costing up to $3,000. "Each move in the game requires complex calculations, resulting in substantial energy consumption. In contrast, human brains function differently, he said.

"So embedding the scientific principles of human thought into computers represents a crucial and foundational aspect of AI's future development. However, this endeavor demands thorough basic research and won't achieve instant success. It's a gradual, step-by-step journey."

Video edited by Zhong Xia, Zhao Chenchen

Cover image by Jia Jieqiong, Liu Shaozhen