Source: Content compiled from snexplores

Inside the new computer, a stream of tiny LEDs glows green. Those lights have a job of their own. They're doing calculations. Right now, those math operations are telling the computer how to recognize images of handwritten numbers.

The computer is part of a Microsoft research project.

The device has no case. "You can stick your hand inside the computer and block the light," says Hitesh Ballani. Once he does that, he explains, "it doesn't know what problem it's solving."

Suddenly, the computer no longer speaks numbers correctly, but spits out "random guesses," says Ballani. Then he takes his hand away. Now the stream of light calculates correctly again. "It's really satisfying to see it," he says.

Ballani is a computer scientist at Microsoft Research in Cambridge, England. His job is to develop this machine, the Analog Optical Computer. As the name implies, it uses light to do calculations. Analog means it does the opposite of digital.

In a digital computer, a signal is either 1 or 0. It's like a light switch that can only be turned on or off. An analog signal is more like a dimmer switch. Any light intensity between fully on and fully off is possible.

To be clear, light is already an essential part of digital computing. Fiber-optic cables carry data back and forth between computers in the form of beams of light. But to actually compute, these computers must first convert light into electricity.

Electronic computers have reached their limits. It's becoming increasingly difficult to increase their speed and power. At the same time, new technologies such as artificial intelligence require more and more computing power. "The demand for computing power is exploding," Barany noted. These systems are also consuming more and more energy.

Analog optical computers and other new optical computing techniques could help meet such demands. The technology could also help green artificial intelligence and other emerging technologies by reducing the energy required.

01 Some simple math

Do you wear glasses? If you do, congratulations! You have your own optical computer.

When light from a scene hits these lenses, it deforms. The scene goes from blurry to sharp. Whatever you look at, the glasses calculate the scene's changes. "And they do this at no energy cost - beyond the original cost of making the glass and bending it," says Charles Rox-Kams. He works on optical computing at Stanford University in Stanford, California, and at the Massachusetts Institute of Technology in Cambridge.

Lens filters also perform simple calculations on light. A clear filter allows all light to pass through. Ballani explains that this is equivalent to multiplying the intensity of the light by one. On the other side, you always get the same intensity.

A black filter blocks all light, so you multiply by zero. A colored filter is like a pair of sunglasses that only blocks some of the light. This is equivalent to "multiplying the intensity of the light by a value between 0 and 1," he explains. You can also add light. In the chip in a smartphone camera, the intensity of many light beams falls on the same area, which brightens that point in the image.

The multiplication or addition takes place over the entire area of the filter or camera chip. Many different light beams change at the same time.

Simple operations on large numbers happen to be a very important part of artificial intelligence. But in the central processing unit (CPU) of a digital computer, all these calculations are usually done one by one. This takes time and energy. Optical computers can perform such mathematical operations simultaneously more easily. Note that some specialized electronic chips used in artificial intelligence, called graphics processing units (GPUs) and tensor processing units (TPUs), can also perform large numbers of mathematical operations simultaneously.

02 Super Smart Lenses

Some new optical computers take the idea of glasses or camera lenses to the next level.

Edogan Ozcan is an engineer at UCLA. His team designed sensors he calls diffractive optical processors. "Think of them as futuristic lenses," he says.

Each processor has multiple layers of glass or other material through which light passes. Each layer contains dozens of tiny structures. Each structure changes the light in a different way. Although they can perform complex tasks, these processors don't require energy to run.

Factories might use it to look for defects in the products they make.

"Let's say I'm making some cancer drug," Ozcan suggests. It's important to find any defective drugs so they don't get to patients. Almost all drugs are defect-free. Yet today's factories use computers to process images of every batch of drugs, scanning for the occasional defect. That wastes time and energy.

"The bottom line is, there's too much data," Ozcan says. "We're drowning in data."

A smart optical processor could change that. It could trigger a camera to automatically detect defects. Then the factory would just have to take a picture of the defective drug.

Ozcan's team designed an optical processor for the task. To test whether it would work in a realistic situation, they used square silicon samples instead of what a factory might produce. Silicon is a semiconductor material used in standard computer chips. Some of these samples had defects etched into them. The engineers had to figure out which tiny structures to add to the layers of the processor so that the processor could see those defects.

To accomplish this, the researchers created a virtual optical processor with random structures in its layers. They also created a virtual simulation of the original silicon block. Then, they added defects to more than 20,000 virtual silicon blocks.

They used machine learning on a regular computer to train their new virtual processor. They showed it virtual samples. At first, it could only guess randomly whether there was a defect. But after each success or failure, the thickness of the processor's many tiny components (which changes the amount of light it lets through) was adjusted to make future correct answers more likely.

In the end, the virtual light had a different intensity when it passed through a defective sample than when it passed through a good sample.

Next, the team had to test whether the design would work in the real world, too. They 3D-printed the processor and tested it with a set of 10 real silicon wafers. The team had already etched defects into nine of them.

The processor correctly identified all the defective wafers and ignored the good ones. Ozcan's team reported this success in a paper published in Nature Communications in October 2023.

The team has also designed optical processors for many other types of tasks. One of their latest processors has layers that can be rotated. This helps it encrypt data.

03 Millions of multiplications

Once the smart lens is 3D-printed, its structure doesn't change. So every new task requires designing a new lens. Ozcan notes that you might only need to reprint one or two of its many layers. But it's not truly programmable like today's typical computers.

Microsoft's analog optical computer, by contrast, can be reprogrammed. But unlike Ozcan's smart lens, it doesn't directly capture the light emitted by an object or scene. Instead, it recognizes handwritten images from electronic digital image files. A component called a modulator converts this electronic data into light by varying the brightness of the system's green LEDs.

Those LEDs now shine onto a chip. It's just like the chip inside a projector that teachers use in a classroom. That projector chip brightens or dims each ray of light to perform the math required for the programming.

"There are 4 million pixels on this chip," Ballani notes. That means "as the light passes through," he says, "you could theoretically do 4 million multiplications [simultaneously]."

Millions of multiplications happen over and over again. The answer to each math problem feeds back into the system to brighten or dim the light in new ways. After the calculation is complete, the computer converts the light back into a digital electronic signal. This becomes the answer to the initial question.

So it is an optical/digital hybrid computer.

The light that is cycled through the LED and projector chips can only perform certain types of calculations. Other types of light are transmitted through the computer's electronic component. This component also works very differently from the CPU in a normal computer. It processes analog electrical signals, not digital ones.

Other companies are also working on ways to make hybrid computers that use both digital electronics and optical technologies.

Lightmatter, based in Mountain View, California, is developing an artificial intelligence chip. It uses lasers to perform light-based multiplication operations. All other types of calculations will be performed by typical electronic chips. Lightelligence, based in Boston, makes a computer processor called Hummingbird. It uses light to quickly transfer data between different electronic components within the processor.

04 Teaching New Skills for Ease

Fully optical general-purpose computers don't exist yet. There's a reason for that. Photons are harder to control than the electrons that drive ordinary computer calculations.

Inside a typical digital computer, components manage the flow of electricity. It's a bit like controlling the flow of traffic on a road. Transistors can quickly start and stop the flow of electricity or increase the rate of current flow. Diodes force the current to flow in only one direction.

Most types of computing require this kind of precise traffic control.

Photons, however, don't follow the same rules. It's hard to make them stop, start, or flow in only one direction. "Most photons don't like to interact with their surroundings at all," says Jennifer Dionne. She's a materials scientist at Stanford University.

Controlling photons usually requires bulky materials, such as large magnets. But computer chips are very small. Without the ability to control the behavior of light at these small scales, "I think we're going to be in the dark ages of computing," Dionne says.

Thankfully, she's an expert in finding new ways to control light. In 2019, her team designed a new type of diode. It's very small. But it can make light flow in only one direction. Dionne made the material for this diode act like a magnet.

"When a photon passes through the material, it feels like it's in a magnetic field," she explains, "so it can only travel in one direction."

Since 2019, her team has been working on materials for switching light intensity. "We've shown that we can turn light on and off very quickly," she says.

Her team has combined it with the new diode. Her doctoral student Hamish Carr Delgado now plans to develop it into a product. It may not be used in optical computers right away. But it could be very useful in data centers, Dion says. There, it could help guide data (which moves in the form of light) between various electronic computer components.

To continue to improve computers in the future, Dion says, we need to take a range of different approaches. Optics is obviously a promising approach. But it's far from the only one. There's also quantum computing. And computers that mimic the way the brain works.

Both types of quantum computers will likely use light as part of some of their operations. For example, PsiQuantum, based in Palo Alto, California, has developed a way to perform quantum operations using single photons. They've built tiny components that can generate and detect single photons. They also make components to control photons.

So far, they have assembled tens of thousands of these components onto silicon wafers the size of dinner plates. When viewed under a microscope, some look "a bit like alien spaceships," says co-founder Pete Shadbolt.

Many new approaches are emerging. Together, they suggest that light could change the future of computing.

View more at EASELINK