Photonic Computer

Photonic computing refers to a computer system that uses optical light to form the basis of logic gates rather than electrical transistors. The promise of photonic computers is that light travels faster than electrons, increasing the potential speeds for compute tasks such as machine learning or other compute-intensive tasks. And if it can be mass-produced at practical sizes, photonic computing holds the promise of fundamentally increasing the speed of everyday computing. Another exploitable property of light is that different wavelengths of light do not interact with each other, which allows them to carry parallel streams of data, with each polarization of light capable of being used as an independent information channel to enable more information to be stored in a single channel to enhance information density.

Photonic computing offers distinctive opportunities in terms of developing computing forward. Beyond AI solutions and machine learning, as noted above, photonic computing offers solutions capable of massively parallel processing required for those types of applications and deep learning neural network computations. This makes photonic computing conducive to various other applications that can be built upon these networks. Applications include the following:

• Autonomous driving
• Predictive and preventative maintenance in manufacturing
• Design of control and vision for robotics
• Recommendation of a product in e-commerce and advertising
• Pharmacy, pathology, and cancer detection
• Digital signal analysis and signal processing
• Language translation and text-to-speech (or language development)

The idea behind photonic computers has been around since at least the 1970s (if not earlier), with optical matrix multiplications first demonstrated. Many roadblocks toward photonic computing have existed that have not been solved on a practical level. For example, many photonic components are not integrated as easily as common transistor-based systems. However, increasing use of fiber optics for communications—such as networking where fiber optic cables have shown superior speed and bandwidth compared to traditional aluminum or copper wires—has led to more research and development into photonics for computing.

A complete photonic computer would work similarly to a traditional computer, using logic gates and binary routines to perform calculations. However, the photons used in photonic computers are generated by LEDs, lasers, or other light-based devices that can be used to encode data in light in a process similar to using electrons in traditional computing. The use of light in computing has been suggested as resulting in the development of exascale computers, which are computers capable of performing billions of calculations every second, or 1000 times faster than current systems. By using the speed of lights, it is possible to achieve this level of processing speed.

There are various benefits of photonic computing, which have led researchers to continue in the development of photonic computers. These include the following:

• Unconstrained: Unlike electrons, which transmit information in traditional electronic computing, photons have the ability to travel in free space, allowing photonic computing to develop three-dimensional and parallel transmission of information.
• No crosstalk: Electrons run the risk of interfering with each other, but because photons are not charged, light beams can pass through one another and remain unaffected.
• Better storage: Optical materials have a greater potential storage density than traditional magnetic or electronic storage materials and offer immediate access independent of when data was stored.
• Less heat: Electrons create heat when they pass through semiconductors, and to avoid overheating, the heat must be removed from computers and electrons, which can further limit the density of the compute components. With photonic computing, the photon's move in free space, meaning devices do not overheat, and the need for cooling is reduced.
• Energy efficient: traditional computers use a lot of energy, estimated to be around 10 percent of the world's energy production; however, photonic computing's data density could dramatically reduce the necessary energy to perform similar computations.
• Faster computations: As photons travel faster than electrons (as they travel at the speed of light), computers and electronics would operate more quickly with photonic computing. And with the data density of photons versus electrons, photonic computers are faster and more efficient than traditional computers.
• AI capabilities: Some believe artificial intelligence will be constrained by electronic computing, and for AI to work optimally, it will require the ability to perform near-immediate data access and computations, both of which can be achieved with photonic computing.

When attempting to use light for computing, there are a few technical challenges. The first is size. A switch from electricity to light requires that the photonic computer is a similar size to electronic, which is difficult because, in atomic terms, light is large and creates a constraint when trying to use techniques to bend light inside microchips and processors. Some work has been done in solving that issue, by using surface plasmons, which have a problem with power. Plasmons lose power quickly, and to maintain power, there would have to be a wire included that translates more heat as a byproduct of the energy needed to maintain the plasmon. However, some researchers are finding materials that could allow surface plasmons to be transported on the nanoscale.

Perhaps unsurprisingly, photonic computing is being developed by single components capable of being integrated in more traditional compute systems. There are several difficulties in developing photonic computers and components. One such difficulty in developing photonic computing is the non-linearity of traditional transistor chips. The non-linearity allows the transistors to switch on and off, which enables them to be fashioned into a logic gate, and accomplishing this non-linearity is easy with electronics. But in light, the output of an optical device is typically proportional to its inputs, making the photonic devices incredibly linear when compared to traditional transistors. It has also been pointed out that the linearity could work in specific cases, such as in deep learning systems that rely on linear algebra that could be sped up and maintained with a linear photonic system.

Another solution to the problem of linearity, proposed by researcher Ryan Hamerly, is to use beam splitters that allow two perpendicular beams of light to be fired at from opposite directions. Each beam of light is split by allowing half of the light to pass through one side of the beam splitter, and the remaining light is bounced at 90 degrees from its origin. This allows for two inputs and two outputs, which can make a matrix multiplication possible (which is how a traditional computer performs calculations).

Light switchable properties, through beam splitters or in some research through manipulation of existing or new material properties, can be paired with specific polarizations, directions, and light or optical pulses, which can create specific information processing for photonic computing. This could lead to increased computing enhancement when compared to conventional electronic chips, as the computing speeds can be increased and modulated by nanosecond optical pulses.

Where traditional electron-based computers use silicon channels and copper wires to guide and control the movement of electronics, the same is difficult as light does not easily bend, requiring more geographical space to make similar compute routes. However, one proposed solution to achieve a similar effect is using plasmonic nanoparticles. These particles can guide and control photons' movement and allows them to turn corners and continue on a pre-determined path without significant loss of power or conversion to electrons, which would make it possible to create compact and efficient optical computing devices.

A key technology in the field of photonic computing and in the development of more sophisticated photonic computing components is the development of integrated photonics. In this case, rather than developing complete systems or complete components, researchers integrate photonic components into a single device to create a more efficient and scalable approach to computation. This has included running light-based logic gates, which can run a million times faster than conventional electronic logic gates found in traditional computer processors.

These "hybrid" photonic systems—where photonic components can be integrated into a traditional electronic computer—are more likely to be developed, especially in supercomputers, where integrated photonic components can be used for data transfer and gradually enhance or take over specific types of computation.

In 2022, researchers developed a photonic computing processor that utilized polarizations of light. In the processor, photonic computing uses multiple polarizations channels that, as noted above, increased the computing density when compared to traditional computing. This is partially achieved through the development of a hybridized-active-dielectric (HAD) nanowire, which uses a hybrid glassy material that shows switchable material properties upon the illumination of optical pulses. Each nanowire also showed selective responses to a specific polarization direction, allowing information to be simultaneously processed using multiple polarizations in different directions.

Another development in photonic computers has been the integration of photonic components in a quantum computer. One such device, named Borealis, has been developed using integrated photonic components capable of taking 36 microseconds to complete incredibly complex tasks—such as a task anticipated to take a conventional supercomputer more than 9,000 years to complete. The Borealis quantum computer also uses a qubit based on photons that can, in principle, operate at room temperature, reducing the need for cryogenic cooling systems. These photon-based qubits can also be readily integrated into fiber optic-based telecommunication systems.

Borealis also consists of qubits in so-called "squeeze states," consisting of superpositions of multiple photons in a light pulse. The squeezed state can exist in states of 0,1,2,3, or more compared to the traditional qubit in a state known as superposition can symbolize both a 0 and 1 of data. This means the squeeze state can represent more, and Borealis can generate trains of up to 216 pulses of squeezed light. However, this is not equivalent to a 216-qubit traditional device, as the squeezed-state qubits can be used to address different classes of quantum tasks than those traditional systems.

Researchers have also used the Borealis quantum computer on a task known as Gaussian boson sampling, in which a machine analyzes random patches of data. Gaussian boson sampling has practical applications, such as identifying which pairs of molecules are the best fits for each other. However, for some, more importantly, has been a key advance made in Borealis being the use of photon-number-resolving detectors. Prior machines have been designed to distinguish between "photons detected" or "no photons detected." Being able to detect multiple photons exponentially increases the number of computational problems a photonic quantum computer can tackle, with Borealis being able to perform more than 50 million times as fast as previous photonic quantum computers.