Indium phosphide laser arrays have been grown on a 300 mm silicon wafer by Ghent University and imec, the Belgium nano-electronics R&D centre. Growing indium phosphide lasers directly onto the silicon wafer promises compact monolithic silicon photonics circuits.
Silicon photonics chips are hybrid designs because of silicon’s inability to generate light. Silicon photonics companies either couple a discrete laser to a chip or bond indium phosphide wafers or ‘chiplets' to the silicon wafer and process it to create working lasers that become part of the silicon photonics chip. Growing lasers directly on silicon creates a third approach for the densest applications.
“Monolithic integration offers the best scalability once you can grow III-V [material] and do wafer-scale processing,” says Professor Dries Van Thourhout of Ghent University. “But it is also the most challenging to implement in terms of pure physics.”
Depositing indium phosphide on a silicon wafer is challenging because differences in the two crystal materials causes defects. Imec and Ghent University have not stopped such defects but has confined them by depositing indium phosphide in pre-etched v-shaped grooves.
The defects propagate along the v-groove and are confined to a layer 20 nm thick compared to alternative approaches that grow the indium phosphide across a wafer where defects propagate several microns deep. The bulk of the deposited material is of high quality, says Van Thourhout.
The challenge is that the amount of indium phosphide material available overall is far less, since the v-groove slots are 300 to 500 nm wide only. “We have these [narrow] slots and we have to adapt the laser design accordingly,” says Van Thourhout.
Ghent University uses the indium phosphide-deposited wafers made on imec’s 300 mm wafer pilot line and etches gratings on top to create the working lasers.
A 20mW external pump laser is used to get the array to lase, says Van Thourhout, while the output power of each laser in the array is 10 mW. The lasers operate in the 910 nm to 930 nm region.
Monolithic integration offers the best scalability once you can grow III-V [material] and do wafer-scale processing. But it is also the most challenging to implement in terms of pure physics
To get the lasers to work at 1,300 nm and 1,550 nm telecom wavelengths, another material such as indium gallium arsenide will need to be grown on top of the indium phosphide, an area Van Thourhout and his team are investigating.
However, the main challenge still remaining is to use electronic injection to drive the lasers. This requires a PIN junction to be integrated on-chip to inject carriers into the laser, and that will require adding electrical contacts which must not induce optical loss in the laser.
“That will certainly be a design challenge, getting the right doping level and so on,” says Van Thourhout. “We also have to find a way to inject current into the device without disturbing the optical field.” Only then can the reliability of laser array be determined. “That [reliability] is something that at this point is unknown but is very important for any commercial device,” he says.
Ghent University says the optically pumped lasers have not shown any breakdown and almost all the devices tested are operating well but he admits that the work remains preliminary.
Ghent University says the advent of monolithic lasers will complement existing discrete laser and hybrid techniques rather than replace them.
“The main target for monolithic is high-volume applications and more integrated designs,” says Van Thourhout. One example is optical links between a CPU and memory. Such designs that integrate optics with ICs will have to be very cheap. “The only way to make something like this very cheap is by a very high degree of integration,” says Van Thourhout.
Meanwhile, Imec has a R&D programme on optical I/O with key partners of its core CMOS programmes. Huawei has been one known partner but others include GlobalFoundries, Intel, Micron, Panasonic, Qualcomm, Samsung, SK Hynix, Sony and TSMC.