As artificial intelligence, cloud computing, and high-performance data centers continue to drive global demand for faster computing, researchers are increasingly looking beyond conventional electronics.
A team at the Massachusetts Institute of Technology (MIT) believes the answer may lie in more efficient integration of electronic and photonic chips. A challenge that has long slowed the adoption of next-generation optical computing.
Through its FUTUR-IC research program, MIT has unveiled a series of advances that could help future microchips transmit data at more than one petabit per second while consuming significantly less energy. The work centers on new devices that simplify the integration of electronics, which process information using electricity, with photonics, which transmit information using light.
According to the researchers, the technologies could also be manufactured using existing semiconductor production equipment, making them more practical for large-scale adoption.
Solving one of silicon photonics’ biggest bottlenecks
For years, engineers have viewed co-packaged optics as one of the most promising ways to improve data transfer inside servers and high-performance computing systems.
Optical communication consumes far less energy than electrical interconnects, making it increasingly attractive as data centers expand to support AI workloads and cloud services. However, integrating photonic chips with conventional electronic processors has remained both technically difficult and expensive.
MIT’s FUTUR-IC program aims to address that challenge by developing components that simplify optical packaging. Among the latest advances are two new optical couplers, an evanescent coupler and a graded index (GRIN) coupler, designed to transfer light between photonic devices more efficiently.
The team also highlighted a third coupler previously developed by researchers led by Professor Juejun Hu.
Together, the devices represent what researchers describe as the first optical equivalents of “solder bumps,” the tiny metallic connections that link today’s electronic chips. Instead of carrying electrical signals, these optical connections transfer light between photonic components, potentially making future electronic-photonic packages easier to assemble and manufacture.
Why photonics matters
Unlike electrical signals, which encounter increasing resistance and power losses as data rates climb, optical communication can transmit enormous amounts of information with significantly lower energy consumption.
According to FUTUR-IC director Anu Agarwal, the program’s long-term objective is to move data transmission from today’s hundreds of terabits per second to beyond one petabit per second. The research group argues that using electronics primarily for computation while relying on photonics for communication could substantially reduce the energy demands of future computing infrastructure.
The need is becoming increasingly urgent. As AI models grow larger and cloud services continue expanding, data centers are expected to consume an ever-greater share of global electricity. Photonic integration is widely viewed across the semiconductor industry as one of the most promising approaches to improving bandwidth without a proportional increase in power consumption.
Different couplers for different applications
Rather than developing a single universal solution, the MIT researchers created multiple optical coupling approaches optimized for different requirements. The GRIN coupler offers broader wavelength compatibility, allowing it to operate across a wider range of optical signals.
The evanescent coupler, meanwhile, is easier to fabricate and can be packed more densely, making it suitable for applications requiring large numbers of optical connections within a limited area. According to the researchers, future electronic-photonic systems will likely require multiple coupling technologies, each balancing manufacturing complexity, optical efficiency, and integration density differently.
Beyond chip design
FUTUR-IC extends beyond semiconductor hardware. The program has also introduced Earthster, a modeling platform that helps companies evaluate the environmental impact of semiconductor manufacturing by identifying energy use, material consumption, and carbon-emission hotspots across their products.
In parallel, the initiative is developing workforce training through online courses, boot camps, and educational resources focused on semiconductor resource efficiency.
Although commercial deployment of these technologies remains some distance away, the research addresses one of the industry’s most persistent challenges: efficiently integrating photonics with conventional electronics.