Erbium atoms in a lithium niobate chip hold several photons for over a microsecond

Photons in a quantum processor or network often have to be paused while slower operations catch up ^[12]. The chips that do the pausing can be as small as a centimeter, a distance light crosses in a few trillionths of a second, according to the University of Illinois account ^[1]. A microsecond is a million picoseconds. Holding a photon past that mark means keeping it for something like 200,000 to 500,000 chip crossings ^[13].

Stretching the light path does not get there. Matter absorbs photons, and each extra length of travel raises the chance a given photon is lost ^[2]. Highly coherent atoms can hold light for longer, but fitting them onto scalable nanophotonic platforms has been largely out of reach ^[3]. "Long delays are an outstanding problem in quantum information processing," said Daren Chen, a physics graduate student and co-first author of the paper ^[10].

I like this design for its economy. The erbium atoms that store the light are doped into the lithium niobate waveguide that carries it ^[4]. A tunable laser sorts those atoms by frequency into evenly spaced peaks, an atomic frequency comb, and the comb lets them catch an incoming photon and hold it for a predetermined time ^[5].

The manufacturing argument rests on the host material ^[8]. Goldschmidt made that argument herself. "This is not a one-off bespoke device," she said ^[15]. "Its promise is in its simplicity: we've taken something that can typically only be done in a highly specialized laboratory environment and recast it in a platform that can be made commercially at scale by people who don't know anything about quantum optics," she said ^[16]. Priyash Barya, an electrical engineering graduate student and the other co-first author, said: "No one else has stored light on a chip in a platform like this, with this potential for scalability" ^[17].

Potential is the accurate word. The account calls the device easy to replicate ^[18]. A fabrication problem, in the sense a foundry means it, is about yield and device-to-device spread across a wafer. The account does not report how many devices were built, the retrieval efficiency, the number behind "high-fidelity", or how many photons were held at once ^[6][14].

Retrieval efficiency heads the group's own list of what must improve ^[9]. It is the figure an integrator would ask for first. A memory that hands back only a small share of the photons it takes in adds loss to the link it was meant to help.

I think the fair reading is that the host is ready for fabrication and the memory built in it is still a research device. If the planned work raises efficiency and storage time, scaling up would use thin-film lithium niobate processing that already exists at wafer scale ^[8][9]. "This project is one component in the set of things we're working on," Goldschmidt said ^[11].