黑料福利网

Eindhoven University of Technology (黑料福利网) is globally recognized as one of the leaders in photonics research and played a pioneering role in the development of photonic chips. 黑料福利网 has been at the forefront of photonics since the 1980s, co-founding the well-known COBRA Research Institute in 1994, which evolved throughout the decades since then, with photonics research playing a big role in the new Casimir Institute today. While photonics has enabled major advances in high-bandwidth communications and optical sensing across industrial and healthcare sectors, optical processors have yet to find their way into practical applications. One critical building block, a photonic memory device, is one of the missing links.

Motion Imager鈥檚 experience in device, system design and its integration capability combined with a clear roadmap to industrialization complements 黑料福利网鈥檚 photonics expertise in device and chip research aimed at foundry scale manufacturing. This can bring an inflection point for future photonic processors. Photonic processors combined with photonic memory show great promise in mitigating many of the today鈥檚 challenges and limitations of state-of-the art silicon electronic processors or photonic accelerators with electronic memory technology. 

The energy penalty in electronics memory is a physics problem

Physics sets limits to electronic memory. In traditional silicon memory (CMOS), we don't just move data, we fight physics. Because electrons have mass and charge, their motion through a semiconductor involves scattering that converts electrical energy into heat, resulting in Joule heating.

As we push for higher speeds, this heating drastically increases, leading to high inefficiencies and wasted energy. In modern data centers, nearly 30% of primary energy is used to deal with heat management, with Joule heating at its origin. This high level of energy consumption represents a substantial environmental burden and a major economic cost, reaching billions of dollars. Next to the heat, there is also a limit to the maximum speed that can be achieved in electrical memory because of resistive-capacitive effects. We are hitting a thermal wall that electrons simply cannot scale.

Photonic memory is not bothered by resistive effects and Joule heating. It is the massless photon that interacts with matter. Motion Imager and Photonic Integration are going to work on a new concept which exploits light-matter interaction to produce states and phases which can be used to encode and retain information, forming a photonic memory.  

It is time to break the binary silence  

In traditional silicon architecture, memory is a passive binary bin that holds a charge (1) or doesn't (0). This necessitates the constant, energy-expensive shuffling of data from memory to the CPU. This memory wall is the primary cause of the Von Neumann bottleneck. According to various estimates, around 60% to 70% of the energy is lost in just moving the data between the memory and the computing unit.

The project envisions the photonic memory as a multi-modal device for multilevel encoding and to perform additional compute task within the memory unit. Unlike electronic bits, these photonic states leverage the unique degrees of freedom of light that can encode multiple bits of information, leading to increased energy efficiencies in the order of tens of femtojoules per transferred bit, more than 1000X improvement than conventional memory. 

Unlocking photonics for next-gen compute

Current state-of-the-art data center infrastructures built around CPUs and GPUs face fundamental memory and interconnect bottlenecks that undermine both environmental sustainability and economic efficiency. This research project focuses on the photonic memory which can be subsequently embedded into a memory stack architecture of Motion Imager for their ultra-high throughput compute platform. This industrial collaboration could play a pivotal role in aligning the integrated photonics value chain for compute-centric applications.