Moore’s law dictates that the number of transistors in an integrated circuit will increase approximately every two years, along with computer power. This principle has been largely correct since 1965. However, one can assume that Gordon E. Moore, the co-founder of Intel, never considered that chips would reach their miniaturization limitations, causing computational power to lag.
Thankfully, a team of engineers from the University of California has developed a novel method that utilizes photonics to provide more computational power without sacrificing size. Here’s what you need to know.
Shrink’em
The world has been on a quest to shrink computers since their invention. Some of your grandparents might even remember when computers required an entire room and many people to operate. Today, your smartwatch provides more computational power than these behemoths, and that’s just the tip of the iceberg in terms of microcomputing. The journey from an entire room down to today’s microelectronics has been an exciting one, filled with experimentation, discovery, and sometimes failures.
Physical and Cost Limitations
Notably, there’s a point where manufacturing such microscopic components is cost-prohibitive. According to most researchers, the market has hit this point. Additionally, the computational gains are not in line with Moore’s law as the smaller chips can’t provide computing power proportional to their larger counterparts.
As such, there’s a growing gap between available computational power and what’s needed, leading some well-known players in the market, like Nvidia’s CEO, Jensen Huang, to state that Moore’s law is dead, listing application-specific processors as the future.
Demand for High-Powered Computers
The sudden growth of artificial intelligence and machine learning systems has increased the demand for powerful computing and cloud platforms. These demands have outpaced the chip design in terms of performance, creating a bottleneck in the AI development sector that limits innovation. Now, the race is on to provide a solution that can keep in line with the demands of today’s high-powered computer and AI systems.
Computing Power Solutions
Some analysts believe that one solution is to create special chips that have the logic and processing on the same chip. This approach does help to reduce latency and energy usage and improves performance. However, computational limitations still make it unideal when discussing massive AI data requirements.
Memory Computing
Another solution that has researchers excited is memory computing. This method of storage utilizes a fast-access ram instead of a spinning hard drive. Specifically, RAM from across a network of computers is set up using middleware software to run the memory in parallel. This method of storage is 5000x faster than traditional methods but cannot still keep up with Moore’s law.
Photonics
Photonics is another method of high-powered computing that has researchers interested. Photonics functions by detecting light waves using electronics. The science revolves around generating and controlling light as it passes through a matrix of programmable optical weights.
These weights integrate a 2D array of non-volatile optical modulators that enable it to perform a linear transformation on a vector of optical inputs. This strategy provides faster switching speeds than RAM computing methods.
Current Limitations to Photonics
Some of the disadvantages of photonics in computing include that the current systems can only be rewritten about 1000x. This limited lifespan makes them an expensive option. Additionally, the optical weights have low storage density and programming is slower than traditional chips.
Photonic computing systems come in various forms and designs. However, these systems require specialized manufacturing processes that are far more expensive than their counterparts. Despite the cost restraints, many researchers believe that photonics is the future of computing and that Wright’s law is the key to its implementation.
Wright’s Law
There is also Wright’s law, which will continue to play a role in our ability to advance semiconductors. This law is a manufacturing principle that was put forth by aeronautical engineer, Theodore Paul Wright. While working in an aircraft manufacturing plant, he noticed that labor requirements decreased by 15% as manufacturing processes improved, even as production levels increased.
These cost savings came from improved processes, technology, recovery systems, and other upgrades made during production. As such, many analysts believe that Wright’s law will help to make photonics a more available and less expensive option in the future.
You can see Wright’s Law already taking place in the AI sector. A few years ago, it would have been impossible for the average person to access or operate an AI system. Creating AI models, managing, and upgrading them was just too expensive. Additionally, no one had created a reliable way for the AI to interact with humans.
However, since programs like ChatGPT hit the market, anyone can now utilize these powerful tools to improve their efficiency and creativity. These systems integrate large language models that make it simple for anyone to operate them from simple chat prompts. This development drove AI adoption through the roof, leading to the current computational power shortages.
Photonics Study
The study “Integrated non-reciprocal magneto-optics with ultra-high endurance for photonic in-memory computing“1 published in the journal Nature Photonics reveals a new method of optical computing that could revolutionize the market. The study delves into encoding optical weights for in-memory photonic computing in detail.
Source – Nature
The researchers chose to use magneto-optic memory cells and a resonance-based photonic architecture to accomplish their computational goals. The strategy relies on non-reciprocal phase shift in magneto-optical materials as a means to implement photonic in-memory computing.
A new mathematical model enabled the team to test magneto-optical materials. They found that cerium-substituted yttrium iron garnet (YIG) enabled them to utilize an external magnetic field to control the propagation of light. This controlled light could then be used to conduct calculations.
The study demonstrated how tiny magnets can store data efficiently and access it lightning fast.
The process of encoding the magnets works by setting the unit’s magnetic domain strength. To operate the device, electronic signals must be converted to the optical domain using electro-optic (E/O) modulators for programming and storage. From there, Balanced photodetectors (BPDs) convert the differential optical signal back to an electrical signal so that the CMOS logic can process it alongside the SRAM.
Photonics Test
To test their non-volatile magneto-optical memory cell the researcher set up various approaches. They started with a 2-bit electrical input that enabled them to achieve two positive and two negative optical weights. From there, the magnetic field was manipulated to monitor changes.
Cerium-substituted Yttrium Iron Garnet (YIG),
A 500-nm-thick single-crystalline YIG was lab-grown using a radio frequency sputtering method at 750 °C for the testing phase. Specifically, the team decided to use a wafer with a ring radius of 35 µm. Also, a 10-nm-thin silicon oxide layer was integrated as a way to separate the silicon layer from the Ce: YIG layer.
Photonics Programming
Programming the state of the memory cell requires a radial in-plane magnetic field supplied by an integrated gold electromagnet. The system can electronically measure the optical loss changes depending on the direction and magnetic field applied via ferromagnetic thin film.
This arrangement creates a programmable, non-volatile magnetic field, allowing engineers to induce a non-reciprocal optical phase shift in the memory cell. Notably, the study documented both clockwise (CW) and counterclockwise (CCW) modes of a micro-ring resonator as a means to program and access computational data.
Photonics Life Cycle
Testing the lifecycle of the memory was another step. The team programmed an arbitrary function generator that cycled between write and erase pulses as part of their approach. The system was adjusted at a rate of 10 kHz. The engineers then used an amplitude of ±5 V and a pulse width of 500 ns to replicate real-world rewriting. The results were eye-opening.
Photonics Results
The test results demonstrated how this new method of photonics could change computing forever. For one, the new system showed near-limitless rewrite capabilities. Specifically, 2.4B programming cycles were achieved.
Additionally, the engineers determined that utilizing non-reciprocal phase shifts in magneto-optic materials allows them to be deterministically programmed quickly and efficiently. The data showed that a programming speed of ~1 GHz was possible. Additionally, the team achieved non-volatility, multi-level encoding that surpasses current methods. Consequently, this research could revolutionize the computer market moving forward.
Photonics Benefits
Several benefits make the photonics study a game changer. For one, the new system can perform complex operations that require massive computational power. These systems are ideal for AI and ML operations that require matrix-vector multiplication and other advanced sciences.
Reduced Latency
The scientists found that photonics provides higher speeds compared to traditional options. The testing phase revealed 1 ns memory access. This rate is 100x faster than previous photonics devices, opening the door for a new level of innovation.
Lower Energy Consumption
Energy efficiency is another benefit that the photonics system brings to the market. This system can reprogram and access memory at a fraction of the power requirements of other options. The team captured an efficiency of 143 fJ per bit, placing the new system’s requirements at 1/10 of other photonics options.
Reprogrammable
One of the biggest advantages of photonics research is the revelation that these devices have a nearly endless re-programmability cycle. No other computing storage options offer users the ability to rewrite data +2Bx. As such, this scientific research could have a resounding effect on data centers.
Photonics Researchers
The photonics research was led by Santa Barbara, John Bowers, and Galan Moody. Paolo Pintus, Nathan Youngblood, Yuya Shoji, and Mario Dumont also played crucial roles in the research and development of the photonics system. Now, the team will seek to expand their research into other materials to find the best option to power tomorrow’s AI revolution.
Companies that Can Benefit from the Photonics Research
There are a variety of companies that could utilize this research to improve their products and services. Cloud computing networks and data centers are two obvious sectors that will see major upswings in revenue if they can implement this tech. Here is one firm that is perfectly positioned to leverage this information.
Snowflake (SNOW +1.83%) entered the cloud computing market in 2012. It’s headquartered in Montana and was co-founded by Benoît Dageville, Thierry Cruanes, and Marcin Żukowski to provide high-performance cloud computing to the market. Today, Snowflake plays a crucial role in enabling firms to store, migrate, and process data on the cloud.
Snowflake is seen by many as a top stock pick due to its commitment to innovation, cutting-edge services, and positioning. The company recently made headlines due to its One Million Minds + One Platform initiative which seeks to upskill millions of workers on Aiby 2029.
Snowflake Inc. (SNOW +1.83%)
Snowflake is one of the most reputable names in the cloud computing sector. Currently, it handles +4.2B queries daily and has +10,000 customers, including 800 companies that have made the Forbes lists. This support and customer base could drive interest in the company’s shares.
If Snowflake were able to implement a photonics storage system in a data center, they could see massive ROIs. The system would take up less space, require less energy, and produce far less heat. Additionally, they would have an unlimited rewrite lifespan while being able to offer less latency to their clients. In turn, this could drive revenue up.
Photonics – using Light to Take Computing to the Next Level
Photonics could be what the world needs to unlock a new level of computational understanding. This technology offers the longevity and sustainability that engineers seek today. Consequently, many see this research as crucial to the AI movement. All of these factors mean that the Photonics computing movement is right on time.
Learn about other cool computing projects here.
Study Reference:
1. Pintus, P., Dumont, M., Shah, V., et al. (2025). Integrated non-reciprocal magneto-optics with ultra-high endurance for photonic in-memory computing. Nature Photonics, 19(1), 54–62. https://doi.org/10.1038/s41566-024-01549-1
