Europe’s semiconductor leaders are racing to meet energy demands

At last week’s Leti Innovation Days (LID) 2025 event in Grenoble, power consumption was on everyone’s mind.  

With artificial intelligence (AI) driving insatiable demand for compute, datacentres are ballooning in size and energy use. “It’s expected that some datacentres will soon reach 500 megawatts,” said Jean-René Lèquepeys, chief technology officer (CTO) of CEA-Leti, in an exclusive interview with Computer Weekly. “In places like Ireland, where 40% of the country’s electricity is consumed by datacentres, this raises serious questions about sustainability.”

This year’s event, which drew more than 1,000 attendees and featured over 130 talks, revolved around the future of sustainable electronics. In his plenary speech, Sébastien Dauvé, CEO of CEA-Leti, underscored the urgency of tackling energy inefficiencies. “AI is not optional,” he said. “It’s a strategic imperative. But we must not allow it to become a runaway train.” 

Many companies are heeding the call. Representatives from across the semiconductor ecosystem gathered at the event to share solutions – from advanced packaging to photonic interconnects, and new methods to tame the heat and power demands of high-performance computing. Among the speakers, Thibault Basquin of Ardian Semiconductor highlighted the private investment community’s growing commitment. “We’re doubling down on companies building the next generation of power-saving technologies,” he said. “Public-private investment is critical at this inflection point.” 

Lèquepeys announced the launch of Resolve, an ambitious initiative targeting a 1,000-fold improvement in energy efficiency by 2032. “We’re working on 15 technology tracks across three layers of the compute stack, from frugal data use to hardware-level power optimisations.” Backed by a growing consortium that includes Fraunhofer, Silicon Austria and Finland’s VTT, Resolve positions Europe as a central actor in the global sustainability race. 

Lèquepeys also revealed that CEA-Leti is working with partners like Schneider Electric and startups like Ferroelectric Memory Company and Soitec to industrialise early results from pilot lines. These collaborations are already shaping commercial strategies, from wide-band-gap power devices for electric vehicles to photonic links and power converters tailored to AI-centric datacentres. 

Wide-band-gap power devices refer to semiconductors made from materials like gallium nitride or silicon carbide, which can operate at higher voltages, frequencies and temperatures than traditional silicon-based components. These characteristics make them highly efficient for converting and managing power, especially in electric vehicles and datacentres. Photonic links, on the other hand, use light instead of electricity to transmit data between chips or systems. This reduces latency and energy consumption in high-speed communication, which is critical for handling the massive data flows in AI applications. 

Solutions born of frugality and advanced integration

For all the hype around quantum and neuromorphic computing, the near-term wins discussed at LID 2025 focused on pragmatic, scalable innovations. As Deirdre Hanford, CEO of Natcast, put it during her plenary presentation, “We must balance the pursuit of revolutionary technologies with actionable steps that reduce power and increase performance today.” 

Lèquepeys shared that one such actionable step is 3D sequential integration – a method of stacking transistors and interconnects that drastically reduces energy use. “We’ve managed to achieve 3D integration at 450°C, compared to traditional processes at 1,000°C,” he told Computer Weekly. “This significantly cuts manufacturing energy costs while maintaining transistor performance.”

Leti also showcased its Genesis project, a 58-partner European initiative developing PFAS-free, low-emission chip manufacturing methods. “We must be frugal not only with data and power, but also with water and materials,” said Lèquepeys.

He noted that only 15% of electronic waste is currently recycled, despite the high value of the metals involved. “Genesis aims to change this by transforming how chips are made – cleaner, safer, and more recyclable,” said Lèquepeys.

AI is not just the problem – it’s also the solution. Leti is deploying AI tools to optimise fab operations and shorten time-to-market. “Today, too many engineering decisions rely solely on human intuition,” he said. “By embedding AI into process flows, we can avoid costly trial-and-error and accelerate production timelines.”

Lèquepeys also pointed to upcoming applications of Bayesian algorithms and neuromorphic architectures that may dramatically reduce energy for edge inference.

Bayesian algorithms are probabilistic models that update their predictions as new data becomes available. Their adaptive nature makes them particularly effective for edge devices where computational resources are limited, helping to balance accuracy with energy efficiency.

Neuromorphic architectures mimic the structure and function of the human brain using spiking neurons and synapses. These architectures excel at pattern recognition and require far less power than traditional processors, making them ideal for tasks like voice or image recognition on battery-powered devices. 

And chiplets – modular silicon blocks – remain an enticing yet elusive goal. Standardisation is underway, but broader adoption hinges on developing a common library of reusable chiplet functions. “Only vertically integrated players like AMD can fully realise chiplets today,” said Lèquepeys. “But if we want carmakers or smaller OEMs [original equipment manufacturers] to benefit, someone must create the building blocks they can assemble without reinventing the wheel.”

Leti is also investing in next-generation computing paradigms. From analog computing using resistive RAM to photonic AI co-processors and spin-based quantum computing, the institute is laying groundwork for future breakthroughs. But the message from LID 2025 was clear: the biggest gains in AI sustainability will come from smart integration of existing technologies – now, not in a speculative future. 

At a moment when AI’s promise risks being undercut by its hunger for power, Leti Innovation Days 2025 made the case for a European model rooted in frugality, openness and collaborative innovation. As Lèquepeys summed up, “Energy efficiency has always been a part of Leti’s DNA. Now it must become part of the entire industry’s strategy.”