How Thin Film Lithium Niobate (TFLN) Will Power the Next Generation of Coherent Ising Machines – Quantum Computing Report

Although a large part of the articles in GQI Quantum Computing Report As we cover gate-based quantum computers, we must remind our readers that these are not the only game in town. Notably, there are also many other types of physics-based processors that have the potential to outperform classical or gate-based computers on problems using combinatorial optimization. They include quantum annealing, simulated branching machines, Lightsolver, and Coherent Ising Machine. GQI believes that combinatorial optimization problems may account for up to a third of the total quantum applications market. So while these optimization-focused machines may not be as general as gate-based systems, there is still a significant market to make the development of these special-purpose processors worthwhile.

In the past, we reported on research to develop Coherent Ising Machines (CIM) being carried out at NTT Research with several other Japanese and North American academic partners. (See our previous articles here and here). This device works by sending pulses of photons through an optical loop and then using controlled feedback between an output coupler and an injection coupler to provide constructive or destructive interference to amplify the pulses with the correct solution and attenuate. others. This CIM architecture provides full connections of photon pulses and can potentially extend to hundreds of thousands or even millions of photons. The first work on this concept was done at Stanford University.

NTT’s previous CIM implementation used 1 km of fiber optic cable as shown in the image below.

Although this implementation worked, like many first-generation devices, it was large, expensive, slow, and unstable. In a device like this, temperature changes in the room can affect performance and reliability. So the important development that NTT Research is undertaking now is to implement the CIM device with a chip-level implementation using a thin film lithium niobate (TFLN) technology that the company has been developing. In addition to the loop itself, other CIM components can be integrated on the same chip. In the same way that classical computers improved significantly when the technology moved from vacuum tubes to integrated circuits, similar things will happen with the Coherent Ising Machines when they introduce us to the TFLN chips. The resulting system will be much smaller, faster, more reliable, cheaper and easier to build because it will rely heavily on semiconductor manufacturing technology.

Below is a conceptual diagram of how the photon loop would be implemented on a TFLN chip.

Although bulk lithium niobate (LiNb03) has been around for a long time and is used in a variety of telecommunication applications, thin film is relatively new and holds great promise in many different photonic applications due to its wide optical transparency window, low optical losses and appreciable nonlinear coefficients and adaptability to nanofabrication processes. Historically, silicon on insulator (SOI) or silicon nitride have been used for photonics-based quantum computers, but TFLN’s superior characteristics may ultimately change that. NTT is also investigating the use of TFLN for photonic neural network processors optimized for machine learning applications.

Currently, much of the TFLN wafer supply comes from a company in Jinan, Shandong Province, China, called NANOLN. Therefore, having additional sources of this material in other countries would be helpful for a strong supply chain.

For further reading on NTT’s research on the Coherent Ising Machine and also the development of TFLN, we can refer you to the following sources:

April 17, 2024