Last week’s announcement of the “World’s first hybrid silicon laser,” made me both excited and annoyed. I’ve been following the promise of optical interconnect technology for almost two decades, and have written a few reviews of the technology over the years. Researchers have come up with many interesting schemes for routing light around chips, boards, and computers: these include integrated silicon waveguides, planar waveguides, free-space interconnects, and photonic crystal pathways (I haven’t written about these but know people working on them). These all have tremendous potential in terms of improving on-chip, chip-chip, or board-board communication. Apart from the obvious, applications include one that I am passionate about: full interconnection of brainlike artificial neural networks.
The problem has always been that the kinds of materials that are best for manipulating and, in particular, generating light have not been compatible with silicon processing. The best known of these III-V materials (collectively named for their groups in the periodic table) are gallium arsenide (GaAs) and indium phosphide (InP). So what people did was to grow the devices on the optoelectronic materials and then attach them to the silicon, using various interesting techniques to align and bond them.
Last year, Intel announced another development in this area: an improvement in the ability of silicon to amplify light. Silicon-based lasers are genuinely novel devices, and any progress in this area can be considered a big deal.
However, the recent announcement—and it was just that, there was no peer-reviewed technical paper to go with the press release—was not for a silicon laser, but for what they are calling a hybrid silicon laser. This is essentially a III-V device again, but integrated in a more manufacturable way because (according to their online, not-fully-technical material) the new technique needs no aligment, the bonding layer is very thin, and the silicon makes up part of the (non-light-generating) structure. This is not a first, but definitely a healthy development.
Of course we don’t really know how (or how well) this works because the information we have about it is so sketchy: aimed at journalists (and investors?) rather than those who can technically evaluate the technology. I’m sure it’s good work, but how good, how much of an advance, how genuinely manufacturable the lasers are, is not clear. Most important developments are first reported by scientists and engineers in journals like Applied Physics Letters, Optics Letters, or the Journal of Lightwave Technology, or even the mainstream press favorites Science and Nature. Instead, I heard about this story first on NPR and the BBC. The fact that they’ve gone straight to the media, skipping the journals, does not mean that the science is bad. But it does make you wonder what they’ve got to hide. Also, the timing is interesting: usually either a paper or a first demonstration dictates the timing of an announcement. The press material that’s been released does not imply that the first demonstration just happened, and we know there’s no scientific paper, so why now?
I’m still excited. The fact that Intel are into this are in such a big way only bodes well: it means that money is going into an area that I believe may be crucially important to increasing the level of complexity at which computers can operate. But I’m uneasy that a story about a new device architecture and bonding technique, an incremental development that would normally be shunned by the mainstream press (believe me, I’ve tried!) is being covered only because there was some big press event. Further, it’s disappointing that there’s not enough technical material available for me to assess this work for myself and decide whether it’s important enough to pitch my own story on the subject. My fear is that this was the point.
Photo: Hybrid Silicon/InP laser chip from Intel. Photo from Intel Press Release.
Originally posted on Brains and Machines.