From today's NR: Dr. Lucas Soldano, a member of POET's technical staff, is an organizer of a symposium entitled "Hybrid Integration of III-V Devices with Silicon-based Waveguides (Si, SiN, SiO2)", which is also scheduled on September 22, 2022 at 10:45am CEST

A patent that Dr Lucas Soldano co-authored has caught my attention as it is related to some of POET’s current efforts which I am not going to get into right now.

Some good general information is contained in the background.

Planar lightwave circuit active connector

Current Assignee is Broadex Technologies Uk

Abstract

An assembly of waveguide wavelength multiplexers and demultiplexers, together with continuous wave (CW) laser transmitters that interface to grating couplers on a silicon photonics chip, providing CW sources, multiplexed output and optionally multiplexed input, all using a single photonic lightwave circuit (PLC).

My comment:

So a common approach to getting light into a silicon photonics PIC is through a grating coupler at the surface of the chip. POET has perfected the vertical mirror which has extremely high coupling efficiency for getting light into the receive side. As we know POET’s optical interposer can provide waveguides in multiple levels with crossovers to allow the receive circuit and the transmit circuit to co-exist  on top of each other. A unique feature of the POET optical interposer that allows for the smallest footprint.  

Below is the background of the invention which I think can be understood by the forum and is worth the effort to read as it relates to issues surrounding silicon photonics. The patent itself allows for an approach to connecting a continuous wave laser to a silicon modulator through a PLC. Remember that POET is using DML lasers for pluggables so this does not apply to those applications. POET will however be using silicon modulators for co-packaged optics (AI & ML)

BACKGROUND OF THE INVENTION

The present application relates generally to fiber optic communications and, more particularly, to optical communications using silicon photonic chips having optical modulators using silicon interfaced with passive planar lightwave circuits.

My comment: In POET’s platform the optical interposer replaces the planar lightwave circuit with the tightly integrated optical interposer which is the dielectric stack layered directly on the silicon substrate containing the electrical interface with the benefits we have come to know and industry is rapidly beginning to understand. 

Background:

In the past few decades, the speed of electronic processing, powered by increasing levels of integrations and smaller gate geometries has overwhelmed the ability of these same silicon integrated circuits to transmit and receive the information that they process. More and more electrical power and chip real-estate is devoted to driving the higher capacitance lines that carry signals off the integrated circuits. Thus the bottleneck in electronics is frequently the communication between chips, modules, or systems.

At the very longest length scales, telecommunication companies use multi-wavelength communication down a single optical fiber to pack more than a hundred channels, each modulated using various techniques to transport information for thousands of kilometers. The optical line cards and transport systems are complex, large, and expensive, justified by the need for bandwidth efficiency in the very long links that they serve. Currently at the shorter distance scales of a few hundred meters to a few kilometers, the same multi-wavelength approach is used, albeit with a smaller number of channels and simple on-off (NRZ) modulation with more compact transceivers and at lower costs. In both types of multi-wavelength communications, laser sources, usually in Indium Phosphide materials systems generate the light, and the data is then imposed on the signal. In the simplest case, the drive current to the laser is changed to vary the optical output intensity, while in more complex systems a separate modulator receives a continuous optical signal from the laser and acts to vary the intensity or the phase of the light that passes through it. The latter is of course more expensive and complicated, but can be more precise, as a separate modulator can more controllably vary the properties of the light.

My comment: DML lasers have of course progressed in recent  years to allow for higher bandwidth and greater reach. 

It becomes very apparent that POET’s path to 400G, 800G, 1.6T and beyond utilizing flip chipped DML lasers is going to be extremely sought after by industry. Very low cost, high performance, efficient and the smallest footprint industry has seen. The AI and ML applications will be served by the applications utilizing the modulator and this is where I see some very interesting connections to Dr. Lucas Soldano’s work as demonstrated by this patent. Clearly he sees the benefits of POETs platform towards the high density requirements of AI and ML applications.

POET’s announcement of the DML supply source is going to be seen as a major announcement by POET customers and partners and hopefully the market is going to recognize what a big deal this supply represents to industry.

Background of Invention continued

Recently there has been a great deal of excitement in the prospect of using silicon as the material for the modulator. The idea is that the industrial infra-structure that allows the fabrication of complex electronic integrated circuits can be leveraged to fabricate the modulators. Such technology can be useful at all length scales, from complex modulators on the silicon that can create intensity and phase modulation for efficient packing of wavelength channels in very long links (for example DQPSK modulation—Differential Quad Phase Shift Keying, used in long haul links) to simple on-off modulation to code ones and zeros (NRZ-non return to zero) in shorter links.

Perhaps the most significant issue with silicon photonics is that silicon as a material, unlike Indium Phosphide, does not possess a direct bandgap. By that we mean that electrons and holes of the lowest energy have different momentum states, and therefore cannot combine directly to generate light. In a forward biased silicon pn junction, the carriers recombine non-radiatively and thus one cannot make LEDs or lasers in silicon. Generally there have been three workarounds for this problem. The first is obviously to have the light off the chip, so a separate indium phosphide laser generates the light and the light is then coupled to the silicon chip where it is modulated and then sent out. The challenge here is of course the complexity of getting the light on and off the silicon chip, especially if multiple wavelengths or multiple sources of light are needed. The second more ambitious way is to try to incorporate the direct gap indium phosphide material on the silicon. The different lattice constant, chemistry, and processing requirements of the indium phosphide make it difficult to fabricate efficient lasers this way. Furthermore, it is impossible to test or burn-in the laser prior to assembly and the relatively poor yield of the lasers increases the cost of the entire assembly. Perhaps the ultimate solution is to try to make the silicon direct gap by adding impurities or changing the crystal through physical deformation. Needless to say, this is very challenging.

A second related issue with silicon photonics is the challenge of coupling light in and out of the chip. Even if the light-source can be integrated into the silicon, one still requires the light to exit the chip and enter an optical fiber. Silicon modulators typically use extremely small and high contrast waveguides. The core is usually made of silicon that is a few hundred nanometers in scale, and the cladding is typically silicon dioxide with a very low refractive index compared to the silicon core (1.46 vs 3.6). Thus the light is highly concentrated in a very tight waveguide. The high contrast has the advantage of being able to make tight waveguide turns, the light paths almost having the geometries of electrical wires, but also has the disadvantage of being completely mismatched to a mode in a glass optical fiber, where the contrast is typically much less than 1% between the core and the cladding. Grating couplers are frequently used to help with the alignment, but grating couplers generally work only at one wavelength and therefore limit the coupling to a single channel per port.