🔬 optics

Energy-efficient programmable integrated photonics via optimized Euler rotations

This paper introduces a geometric framework that optimizes the energy efficiency of programmable integrated photonics by selecting minimum-energy Euler rotation trajectories on the Bloch sphere to implement unitary transformations, a method validated across diverse silicon architectures for both classical and quantum applications.

Original authors: Pablo Martínez-Carrasco Romero, Andrés Macho-Ortíz, José Roberto Rausell-Campo, Francisco Javier Fraile-Peláez, José Capmany Francoy

Published 2026-05-20

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Pablo Martínez-Carrasco Romero, Andrés Macho-Ortíz, José Roberto Rausell-Campo, Francisco Javier Fraile-Peláez, José Capmany Francoy

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you have a giant, high-tech traffic control center for light. This center, called a Programmable Integrated Photonic (PIP) chip, is designed to route and reshape beams of light to perform complex calculations, much like a super-fast computer. These chips are the future of optical computing and quantum technology.

However, there's a problem: running this traffic center is incredibly expensive in terms of energy.

The Problem: The "Heating" Bottleneck

To steer the light, the chip uses tiny switches called phase shifters. Think of these like the knobs on a radio, but instead of changing the station, they twist the light wave to make it go left, right, or straight.

In most current chips, these knobs are heaters. To twist the light, you have to heat up a tiny piece of metal, which takes a lot of electricity. As these chips get bigger and more complex (with thousands of these knobs), the energy bill skyrockets, making it hard to build the massive processors needed for the future.

The Discovery: Finding the Shortest Path

The researchers in this paper found a clever way to cut that energy bill without changing the hardware. They realized that the math behind steering light is like navigating a globe (specifically, a sphere called the "Bloch sphere").

The Analogy: Imagine you are standing at the North Pole and need to get to a specific point on the equator.
- Route A: You could walk straight down a line of longitude. This is a short, direct trip.
- Route B: You could walk in a giant circle around the world, taking the long way around, and then arrive at the same spot.

Both routes get you to the exact same destination (the light ends up in the exact same state), but Route B wastes a huge amount of energy because you walked further.

The paper shows that for every single light-steering task, there are often multiple "routes" (mathematical combinations of angles) to get there. Previous methods didn't care which route they took, often picking the long, energy-wasting one.

The Solution: The "Shortest Path" Algorithm

The team developed a new "GPS" for these light chips. Instead of just calculating how to steer the light, their system calculates the shortest possible path on that globe to get the job done.

They also discovered a hardware trick to make this even better:

The Single-Knob Problem: Some chips use only one heater per switch. This is like being forced to walk only in a clockwise circle. If your destination requires a counter-clockwise turn, you have to walk almost all the way around the world (360 degrees) to get there.
The Two-Knob Fix: The researchers used chips with two independent heaters per switch. This allows the light to be steered in either direction (clockwise or counter-clockwise). This means the system can always choose the direct, short route instead of the long, winding one.

What They Proved

The team didn't just do this on a computer; they built a real silicon chip and tested it.

Small Scale: They tested simple 2-by-2 light switches. They found that by choosing the right "route" and using two heaters, they could save a massive amount of power compared to the old way.
Large Scale: They simulated huge, complex processors (like those needed for Artificial Intelligence or Quantum Computing). They found that as the chips get bigger, the energy savings become enormous—saving hundreds of watts of power for large-scale systems.
Real-World Apps: They showed this works for:
- Neural Networks: The "brains" of AI systems.
- Quantum Gates: The basic building blocks of quantum computers.

The Bottom Line

This paper introduces a new way of thinking about how we control light on a chip. By treating light steering as a geometry problem and always choosing the shortest path, we can make these powerful optical computers much more energy-efficient. This doesn't require building new, expensive hardware; it just requires smarter software to tell the existing hardware how to move the light more efficiently.

In short: They found a way to make the light take the "express lane" instead of the "scenic route," saving a huge amount of electricity in the process.

Technical Summary: Energy-Efficient Programmable Integrated Photonics via Optimized Euler Rotations

Problem Statement
Programmable Integrated Photonics (PIP) has emerged as a critical platform for optical signal processing and computing, enabling reconfigurable $N \times N$ unitary matrix transformations through meshes of tunable interferometers. While these systems offer high bandwidth and low latency, their scalability is increasingly constrained by energy consumption. As the number of building blocks (BBs) grows to hundreds or thousands, the cumulative physical actuation required for phase shifters creates a significant energy bottleneck. Previous optimization strategies have focused on system-level approaches, such as pruning irrelevant components or redistributing phase requirements across the mesh. However, these methods address how transformations are distributed rather than how they are physically implemented. A fundamental degree of freedom remains unexplored: the mathematical degeneracy inherent in the physical implementation of individual $2 \times 2$ unitary transformations. Specifically, equivalent decompositions of the same unitary matrix into basic Euler rotations can follow distinct trajectories on the Bloch sphere, resulting in markedly different total phase accumulations and, consequently, different energy costs.

Methodology
The authors introduce a geometric framework to analyze and minimize energy consumption at the level of the fundamental BB and extend this to arbitrary $N \times N$ matrices.

Geometric Representation: The optical field within a BB is modeled as a state on the Bloch sphere. A $2 \times 2$ unitary transformation is interpreted as a rotation connecting an input state to an output state. This rotation is factorized into a sequence of basic rotations around Cartesian axes ( $\hat{x}$ and $\hat{z}$ ), corresponding to fixed beam splitters and tunable phase shifters, respectively.
Euler Decomposition and Degeneracy: The transfer matrix of a universal BB is decomposed into Euler rotations: $T_{2\times2}(\alpha, \beta, \gamma) = \hat{R}_z(\gamma)\hat{R}_x(-\pi/2)\hat{R}_z(\beta)\hat{R}_x(-\pi/2)\hat{R}_z(\alpha)$ . The authors demonstrate that the mapping between a target unitary matrix and the set of Euler angles $(\alpha, \beta, \gamma)$ is not unique. Multiple equivalent sets of angles can generate the same transformation but follow different geometric paths on the Bloch sphere.
Energy Metric: The total phase accumulation, defined as $|\alpha| + |\beta| + |\gamma|$ , serves as the proxy for energy consumption. The optimization goal is to select the set of Euler angles that minimizes this sum, thereby identifying the shortest rotation trajectory.
Physical Implementation Constraints: The study distinguishes between two hardware configurations:
- Dual-phase-shifter implementation: Using two independently actuated phase shifters in parallel waveguides allows for both positive and negative rotation angles ( $\alpha = \alpha_2 - \alpha_1$ ), enabling access to both clockwise and counter-clockwise rotations on the Bloch sphere.
- Single-phase-shifter implementation: Using a single phase shifter restricts rotations to one sign (e.g., only positive), forcing the use of complementary rotations ( $2\pi - \alpha$ ) for inaccessible directions, which increases total phase accumulation.
Validation: The framework is validated through:
- Experimental: Fabrication of a silicon-on-insulator (SOI) chip implementing a universal $2 \times 2$ processor and a multipurpose hexagonal mesh (Smartlight).
- Numerical: Simulations of high-dimensional matrices ( $N$ up to 500), feedforward neural networks (MNIST classification), and photonic quantum gates (CH, FSim, RZX).

Key Results

2×2 Unitary Processor: Experimental validation on 2,000 random $2 \times 2$ transformations confirmed that selecting the minimum-energy Euler trajectory significantly reduces power consumption. Furthermore, using dual-phase-shifter implementations allowed for even greater savings compared to single-phase-shifter architectures, which are forced into higher-energy complementary rotations.
Scalability: Numerical estimates for high-dimensional matrices show that non-optimal implementations lead to rapidly increasing power consumption as dimensionality grows. For a $500 \times 500$ matrix, the optimized implementation reduces total power consumption by approximately 413 W compared to the highest-power configuration.
Neural Networks: In feedforward neural network simulations (3 hidden layers, 16–256 neurons), the optimized strategy reduced power consumption by factors ranging from $\times 5.5$ to $\times 8.2$ compared to non-optimized configurations, with savings increasing alongside network dimensionality.
Quantum Gates: For two-qubit gate implementations (4×4 unitary matrices), the approach yielded power reductions ranging from $\times 1.4$ to $\times 2.8$ , demonstrating applicability to quantum photonic processors.
Hexagonal Meshes: The optimization trends were preserved in multipurpose hexagonal mesh architectures, confirming the method's applicability beyond dedicated feedforward designs.

Significance and Claims
The paper claims to establish a general route toward more energy-efficient large-scale PIP processors for both classical and quantum applications. By exploiting the mathematical degeneracy of Euler rotations, the authors provide a physically grounded method to minimize energy consumption without altering the underlying circuit topology or requiring new materials. The framework is technology-agnostic regarding the specific phase-shifter mechanism (e.g., thermo-optic, electro-optic, MEMS), provided the hardware can physically access the necessary rotation directions (often requiring dual actuation for single-sign technologies). The work suggests that as PIP processors scale to thousands of building blocks, optimizing the trajectory of individual BBs becomes a critical factor in overcoming energy-density limitations and enabling the next generation of optical computing and signal processing systems.