Each unit of metamaterials used for simulating optical calculations is smaller than the wavelength of the light they are designed to manipulate

The new architecture based on metamaterials provides a promising platform for constructing large-scale production and reprogrammable solutions that can perform computational tasks using light.

The idea of simulating computers - a device that uses continuous variables instead of zero sum ones - may evoke outdated machinery, from mechanical watches to bomb sight devices used in World War II. But emerging technologies, including artificial intelligence, may benefit greatly from this computing method. A promising direction involves analog computers, which use light instead of current to process information. As Nader Engheta from the University of Pennsylvania reported at the March 2024 APS conference, composite media known as metamaterials provide a powerful platform for building simulated optical computers. In recent work, his team demonstrated a metamaterial platform that can be mass-produced and integrated with silicon electronics, as well as a method of building an architecture that can be reprogrammed in real-time to perform different computing tasks. Engheta said that simulation optical computers based on metamaterials may one day perform certain tasks faster and consume less power than traditional computers.

Metamaterials are synthetic materials made by assembling many units, each of which is smaller than the wavelength of the light they are designed to manipulate. They can be customized to display properties that are not present in natural materials, most notably near zero or negative refractive index. These unique characteristics can achieve unique applications, from subwavelength imaging to stealth.

The design flexibility of metamaterials has inspired several groups to explore strategies for transforming them into computers. In 2014, Engheta and collaborators proposed the first set of proposals. Their simulations indicate that metamaterials can perform a series of mathematical operations, including differentiation, integration, and convolution. This method involves using electromagnetic waves as input functions and manipulating them through interactions with metamaterials, so that the output wave corresponds to the required mathematical transformation of the input.

Five years later, Engheta's team implemented this proposal through experiments. When working at microwave wavelengths, their approach involves a metamaterial with multiple input and output ports connected through waveguides in the feedback loop. Experiments have shown that for a given input, the output of the device is a solution to the so-called Fredholm integral equation, which is used in multiple fields such as fluid mechanics, antenna design, and quantum mechanics perturbation theory. In order to select the metamaterial structure that implements the required mathematics, researchers used "reverse design" - an iterative method to solve optimization problems. The resulting metamaterials have a non trivial "Swiss cheese" structure, uneven distribution of small islands with different dielectric properties - pores, polystyrene, and microwave absorbing materials.

Due to the cumbersome and impractical nature of microwaves, several research groups have begun to extend similar concepts to optical frequencies, demonstrating various computational schemes. Most of these demonstrations use sub wavelength thin metamaterial sheets to manipulate the light propagating in free space and transmitted through the sheets. However, metasurface solutions require complex customized manufacturing processes, which limits the potential for large-scale production, Engheta said.

Engheta and his colleagues have now developed an on-chip platform that can overcome these limitations. Unlike metasurface schemes with free space light propagation, the team's metamaterial design guides light through structured waveguides on silicon chips. Researchers have reverse designed and manufactured a micrometer scale chip with a structure reminiscent of their 2019 microwave design: a set of waveguides that feed light into and out of a flat cavity containing metamaterials similar to Swiss cheese. Engheta said that this structure can be simply ordered from commercial foundries. Compared to microwave cousins, optical chips have simpler mathematical operations - they multiply vectors by matrices, which are useful for artificial intelligence tools such as neural networks. To solve the equation, the solution needs to combine feedback waveguides that connect the output and input, as done in microwaves, which is an engineering challenge that the team plans to address in the next generation of chips.

While working in optics, Engheta is using lower frequency principle verification equipment to drive the mathematical capabilities of analog computers. The latest results of the group have added an important new feature: reconfigurability - the ability of equation solvers to reprogram to perform different mathematical operations. This scheme consists of 5 × 5 modules of RF components. Equipment can be reconfigured by controlling the parameters of each component. As a demonstration, the researchers had their machines solve two different problems: finding the roots of polynomial systems and designing the inverse of the execution element structure. Both of these problems are non-stationary, which means they require a series of steps, each with different mathematical operations.

Engheta envisions that this reconfigurable feature can ultimately be extended to silicon photonics chips. One method is to deposit a patterned layer of phase change material on the top of the waveguide of the device. When heated, this material changes its refractive index, thereby affecting the propagation of light in the waveguide and thus affecting the mathematical operators of this propagation encoding.

Engheta said that programmable metamaterial silicon photonic chips will be a blessing for analog optical computing, as they process information at the speed of light, while traditional digital processors require only a small fraction of the energy required to perform millions of operations. "Here, light passes through a waveguide maze, and when it comes out, you get the answer in one breath," he said. Moreover, since photons and electrons are different and do not interact with each other, parallel operations can be performed simultaneously by illuminating different wavelengths of light through the device. More importantly, such devices will have privacy advantages as they do not require intermediate steps to store information in potentially hackable memory, Engheta said.

Source: Laser Net