Deuterated silicon nitride on 8-inch wafers puts a broadband light source inside the fab

The absorption that spoils ordinary silicon nitride comes from silicon-hydrogen bonds left over from the silane the film is grown from, and they sit at the wavelengths telecommunications systems use ^[5]. Vibrational frequencies scale with the mass of the atoms doing the vibrating, so putting deuterium in place of hydrogen moves the absorption rather than eliminating it; Tan says the peak lands in the 2.1 micron region ^[7]. That bargain has a visible edge. The long-wavelength end of the continuum stops roughly 217 nm short of the relocated absorption peak ^[17]. The account available does not say the deuterium band sets that limit, but anyone extending this platform deeper into the mid-infrared is now heading toward an absorption feature instead of away from one.

The spectral claim is at least self-consistent. A span from 587 to 1,883 nm is a frequency ratio of 3.21, or 1.68 octaves ^[18], which matches the reported 1.7 and the statement that light at one end oscillates more than three times faster than at the other ^[10]. The chip also glowed red under an input that was itself invisible ^[11], which is the cheapest available demonstration that infrared went in and visible came out.

The binding constraint here was never breadth, it was thickness. Nonlinear broadening wants a thick film, and the conventional route makes transparency and thickness fight each other: hours at up to 1,200 C to drive off the hydrogen, which builds enough stress into a thick film to crack it and would destroy any transistor sharing the wafer ^[6]. That combination is why silicon nitride supercontinuum has stayed outside standard semiconductor factories ^[12].

Provenance is doing real work in this result. The fabrication sits with A*STAR IME's silicon photonics department, whose head describes the hard part as translating the photonic design into a process that delivers the required optical performance at wafer scale ^[13]. A wafer-scale claim owned by a foundry process group is a different object from the same claim out of a university cleanroom, and it is the reason the 8-inch figure carries weight rather than decoration.

Breadth alone does not buy a frequency comb. The light has to be coherent and repeatable from pulse to pulse ^[15], and the material available reports a coherence measurement at moderate pulse energies without stating the value. That number, at the energies needed for the full span, is what the metrology case rests on. Fiber-based systems already produce the spectrum; the first author frames the unmet need as compact, energy-efficient and integrable sources ^[14], which is a claim about form factor and power, not about optics ^[3].