The two headline numbers sit awkwardly together. The structure swallows roughly 85% of the light that reaches it [8], and converts a peak of 2.3% of incident photons into collected charge [9]. Divide the second by the first and the internal quantum efficiency is about 2.7% [1], which means around 97 of every hundred absorbed photons produce nothing the readout can use [2].
That gap is the useful part of the result, because it tells you which problem got fixed. The group's own diagnosis of the field's ceiling was geometric: the non-equilibrium doping route they use can only make very thin active layers, so earlier hyperdoped silicon simply did not intercept much light [4][5]. Pyramid micro-texturing and a rear gold mirror answer that by folding the optical path back through the thin layer several times [7]. Absorption is now close to the practical maximum. Whatever is eating the other photons happens after they are absorbed, and the account given to Phys.org does not say where they go.
The sensitivity figure should be read in the same spirit. 4x10^10 Jones at 1.5 microns, at room temperature and zero bias, is presented as a record for hyperdoped silicon [10], which is a comparison inside a material class rather than against the incumbent. First author Eric Garcia-Hemme frames the target as cost: SWIR detection today leans on expensive materials such as InGaAs, and putting the function on silicon would drop manufacturing costs and open mass-market uses [3]. No InGaAs detectivity number appears alongside it [11]. So the case being made is economic, and an operator comparing parts still has no photometric ratio to work with.
The manufacturing argument is the strongest thing here and also the least tested. Because the device is still mostly silicon, it is described as compatible with standard CMOS processing and buildable in the same plants that turn out processors and phone camera sensors [12], and it works without the cooling hardware some infrared detectors demand [13]. Both claims matter more than the efficiency number, since they decide whether a SWIR pixel costs cents or tens of dollars. But the demonstrated stack is not bare silicon. It carries surface pyramids and a gold reflector on the back [7], and neither the texturing step nor the metal is costed in what has been published so far.
For now the honest read is a photodiode, not a camera. At a few percent external quantum efficiency [9] you need either a bright scene, a long integration, or an application that tolerates both, and the physics that caps layer thickness [4] rules out the obvious remedy of growing more absorber. The next move has to come from collection, which is where silicon's intermediate-level trick and its carrier lifetime argue with each other. If the Complutense route gets internal efficiency into double digits without a thicker layer, silicon takes work away from III-V SWIR imagers. If it does not, the useful legacy is the light-trapping architecture, which anyone doping silicon for sub-band-gap absorption [6] can borrow tomorrow.
