The first 24-mode universal photonic processor has been created via femtosecond laser writing—the most complex realized to date. Optimized for quantum-dot emission at 925 nm, the device features average insertion losses of just 4.35 dB, enabling direct use in multi-photon quantum experiments. Suspended waveguides and precisely engineered 2D and 3D microstructures dramatically improve thermal insulation and cut dissipated power to below 10 W, requiring only simple thermoelectric cooling. After calibration, the processor executes unitary transformations distributed according to the Haar measure with 99.7% amplitude fidelity. This work establishes femtosecond-laser-written integrated photonics as a scalable, reliable platform for advancing quantum computing, communications, and sensing.
Scaling up quantum computing requires compact, stable, and reprogrammable devices for processing quantum information. Universal photonic processors are programmable integrated circuits capable of performing arbitrary linear optical transformations on many light modes. Such processors could become the foundation for implementing quantum algorithms proposed by Richard Feynman and David Deutsch, as well as for creating entangled states.
The 24-mode optical processor circuit was created using femtosecond laser writing in Corning EAGLE XG glass. Laser pulses are focused at a depth of 35 µm and form single-mode waveguides with losses below 0.3 dB/cm, optimized for a wavelength of 925 nm—the emission of single-photon sources based on quantum dots. For thermo-optic control, chromium micro-heaters are placed near each of the 552 directional couplers; connections are made with copper using two-step lithography with dry photoresist. Thermal isolation is enhanced by deep trenches (60 µm) that form "bridges" of waveguides surrounded by air, dramatically reducing thermal crosstalk. Calibration involved measurements of 30,000 states and machine learning to determine 576 phase shifts and 13,824 thermal coupling coefficients.
After packaging, the total insertion losses averaged 4.35 dB—a record low for devices of this complexity. The micro-heaters demonstrated high stability: during continuous operation for 12 hours, the resistance drift did not exceed 0.005%/h. A model was trained for calibration, accounting for static phase offsets, coupler splitting ratios, and thermal crosstalk; in total, 30,000 unitary transformations were processed. Final verification on 2,000 random matrices chosen according to the Haar measure (uniformly distributed in the space of all unitary matrices) showed an average amplitude fidelity of 99.7%. Low insertion losses minimize decoherence, which is critical for entangled states. Electrical power consumption for all tested transformations remained below 10 W, allowing the processor to be cooled with a simple thermoelectric element.
The demonstrated processor sets a new benchmark for scalable quantum photonic circuits. It proves that femtosecond laser writing enables combining three-dimensional design, low losses, and efficient thermal management with a mode count sufficient for non-trivial quantum tasks. This brings closer the creation of quantum simulators and computers, fulfilling the long-standing dream of John von Neumann for programmable information processing. Successful calibration using machine learning methods paves the way for fast and accurate tuning of even larger circuits without the need for expensive individual debugging, which is important for optimization problems and modeling.
Further development of the technology will follow the path of increasing the number of optical modes to hundreds and thousands, which will require solving calibration and heat dissipation problems. A transition to cryogenic temperatures is likely for interfacing with superconducting electronics and detectors. Integration with single-photon sources and detectors on the same chip will turn the processor into a complete "all-in-one" quantum platform. Moreover, automatic calibration algorithms based on machine learning could become a standard for controlling multi-mode photonic circuits.
The development will impact quantum communications (including quantum key distribution) and quantum sensing, where programmable linear interferometers are required.
Next steps include demonstrating quantum algorithms (e.g., boson sampling) on the 24-mode device and integration with solid-state single-photon emitters on the same glass chip.
The work is directly linked to the problem of scaling quantum computers: how to maintain a low level of decoherence and high control accuracy as the number of qubits (modes) grows. In addition, the complexity of calibrating large systems resonates with fundamental questions of complexity in quantum control theory.
🎯 The glass used in the processors is similar to the material in smartphone screens, but its transparency and thermal stability make it ideal for femtosecond laser writing of waveguides. Interestingly, the depth of the thermal isolation trenches (60 µm) is thinner than a human hair, yet that's what reduces crosstalk by orders of magnitude.
🎬 The idea of a universal optical processor echoes the concept of light-based computers from science fiction, for example, in Neal Stephenson's novel 'Snow Crash', which features holographic interfaces and optical neural networks.