A research team from Singapore, led by Associate Professor Dawn Tan of the Singapore University of Technology and Design (SUTD) and Dr Luo Xianshu, Head of the Silicon Photonics Department at the A*STAR Institute of Microelectronics (A*STAR IME), has developed a low-loss silicon nitride waveguide that generates broadband light on a chip. By replacing hydrogen with its heavier isotope, deuterium, the team fabricated the low-loss SiN waveguide on an eight-inch wafer using a low-temperature process, demonstrating its potential for large-scale manufacturing and integration with CMOS-compatible semiconductor processes.
Published in Optics Express , their paper “ Octave-spanning supercontinuum generation in a wafer-scale, low loss deuterated silicon nitride waveguide ” demonstrated a chip-scale waveguide that stretches infrared laser pulses into a spectrum running from visible red deep into the infrared.
Lasers are prized for their colour purity by emitting light in a single colour, but many of the most demanding technologies want a beam that spans an enormous sweep of the spectrum at once. This so-called supercontinuum light underpins high-resolution medical imaging, precision measurement, and the frequency combs that keep optical clocks ticking without error.
Today’s supercontinuum sources are mostly built around specially engineered optical fibre, but they are bulky, power-hungry, and difficult to shrink onto a chip. The most compatible chip material is silicon nitride, but making it transparent enough demands conditions so extreme that thick films crack. Standard semiconductor factories also cannot accommodate the process.
“The primary gap is the need for compact, energy-efficient, and integrable light sources,” said Yao Wang, PhD student and first author on the paper. “While fibre-based systems are standard, they are bulky and not easily integrated onto chips.”
The trouble with conventional silicon nitride traces back to hydrogen. Films are typically grown from silane gas, leaving behind silicon-hydrogen bonds that absorb light at precisely the wavelengths telecommunications systems use. Purging those bonds requires hours of annealing at up to 1,200 degrees Celsius—a temperature no chip carrying electronic circuitry could survive, and one that builds crippling stress into thick films.
So, the researchers swapped ordinary silane for its deuterated cousin, replacing each hydrogen atom with the heavier isotope deuterium.
“This substitution achieves a physical shift. The absorption peak moves from the telecommunications range to the 2.1 micron region,” explained Assoc Prof Tan. “Because light is no longer being absorbed at the operating wavelength, the high-temperature baking process is no longer necessary.”
The resulting fabrication process is simple. A film that was 800 nanometres thick was deposited in a single step at below 400 degrees Celsius, comfortably within the thermal budget of standard semiconductor lines, across a full eight-inch wafer, which is the format used in mainstream chip manufacturing. The finished waveguides lose just 0.54 decibels of light per centimetre, low enough for a 5.21-centimetre device to sustain the intense optical interactions that spectral broadening demands.
“The key challenge was translating the photonic design into a process that could deliver the required optical performance at wafer scale. By combining SUTD’s nonlinear photonics expertise with A*STAR IME’s process development and fabrication capabilities, we have taken an important step towards scalable manufacturing,” said Dr Luo.
When the researchers fired 500-femtosecond infrared pulses into the waveguide, the spectrum ballooned from a narrow line at 1,555 nanometres to a continuum stretching from 587 to 1,883 nanometres—a span of 1.7 octaves. This means that light at one end of the spectrum oscillates more than three times faster than at the other. Although the input light was infrared and invisible, red light could be seen glowing from the chip.
“That proved the device was effectively converting the infrared input into a spectrum broad enough to reach the visible range,” said Assoc Prof Tan.
Breadth is only half the story. For metrology and frequency comb applications, the light must also be coherent, stable, and repeatable from pulse to pulse. At moderate pulse energies, the team measured an overall spectral coherence above 0.81, indicating good shot-to-shot stability.
At the highest energies, however, a trade-off emerged, where the spectrum grew broader but noisier, as an effect called modulation instability began to scramble the light. Simulations pointed to a practical remedy. The light stays highly coherent for roughly the first centimetre of travel before the noise takes hold. Optimising the waveguide length could allow users to capture a broad spectrum while it is still clean and stable, before the coherence degrades as the pulse propagates further.
That refinement heads the team’s list of next steps, alongside waveguide designs that maximise spectral uniformity and broadness at high power. The larger goal is full integration: pairing these light sources with modulators, detectors, and other components on a single chip to bring supercontinuum light to medical imaging, precision metrology, and optical communications at manufacturing scale.
Optics Express