Inside waveguide photodetectors

What changes in this design

Typically, light enters a waveguide photodetector straight through a flat, perpendicular facet at the tip of the channel. Some of that light inevitably reflects off the entrance rather than being absorbed, and it bounces straight back the way it came, toward the same optical path it arrived from.

This design cuts the entrance at an angle instead, so incoming light gets bent (refracted) as it crosses into the channel and steered toward the p-doped side of the sensing region, rather than traveling straight through. Any light that does reflect off that slanted surface heads off at an angle too, away from the source.

For a reader, this matters because wasted or reflected light in a detector chip means a weaker signal and can even feed unwanted light back into whatever produced it. A design that steers light forward instead of back suggests a detector meant to lose less light and respond to fast-changing signals more reliably.

Light Bouncing Back Where It Started

A beam of light travels down a channel etched into silicon, headed for the diode that is supposed to absorb it and turn it into an electrical signal. Most of it does. A slice of it, though, hits the diode's front edge nearly straight on and bounces back into the laser that fired it. That reflection is called retro-reflection, and it causes trouble two ways: it can destabilize the sending laser, and it caps how fast the whole link can run.

The fix here is simple to describe. Instead of meeting the incoming light head-on, the diode's input facet is cut at an angle, somewhere between 8 and 65 degrees, ideally landing on Brewster's angle, so light refracts sideways into the diode's p-doped region rather than bouncing straight back out. Ranovus, the Ottawa company behind the patent, says the payoff is bandwidth "exceeding 40 GHz compared to approximately 12 GHz for alternative configurations." Filed May 17, 2024. Published this past September.

Ottawa, Funded to Produce Patents

Ranovus has built photonic chips in Ottawa since 2012, and much of the money behind it has come from government rather than customers. Canada's Strategic Innovation Fund gave the company CAD 20 million in 2018 and another 36 million in 2023, folded into a 100 million dollar project the government explicitly expected to generate 40 new patents. In August 2025, Ranovus announced a further Ottawa expansion worth more than 100 million dollars, backed by a 2 million dollar Ontario grant, bringing outsourced manufacturing in-house and adding 125 jobs.

None of that tells you whether this fix was worth inventing on its own merits. It does tell you the company operates inside a pipeline built, in part, to produce filings like this one on schedule. I can't separate, from the outside, how much of this patent's timing answers to physics and how much answers to a funding target.

The Laser That Never Leaves the Package

The photodiode is not an academic exercise. It's a receiver component for what Ranovus actually sells: co-packaged optics, where the electronics that process data and the optics that move it as light sit on one package instead of separate boards. In March 2024, weeks before this patent was filed, Ranovus announced a 6.4 terabit-per-second version of that chip with a laser integrated onto the same package, claiming 4 picojoules per bit and a 50 percent cut in power, footprint, and cost, with the chipmaker MediaTek named as a partner. The angled facet is solving a bandwidth problem inside that specific bet: that a laser living permanently on the package beats one plugged in separately.

A Failure Mode the Industry Chose to Avoid

That bet runs against something hyperscale data centers have already measured. Laser sources rank among the top three failure modes in optical systems, and running one hotter, at 85 degrees Celsius instead of 25, can make it fail ten times sooner - exactly the heat a laser gets from sitting next to switching electronics. Most of the industry has settled on external, swappable laser modules instead: if the laser dies, you pull the module, not the board. Whether Ranovus's integrated laser beats that pattern isn't something I can check from outside; no public failure-rate figures exist for it. What's public is the direction most competitors chose, and it's the opposite one.

Forty Gigahertz Next to a Faster Number

The market this chip sells into can't agree on its own size. One research firm puts co-packaged optics at 121 million dollars in 2025, growing to 764 million by 2031; another puts 2025 at 91 million, growing past 1.9 billion by 2035. Analysts are fighting over timing too: one report argues shipments are being pushed to 2027 and 2029, citing a yield model implying only 19 percent of finished systems pass; a rebuttal points to laser orders at a major chip designer climbing from roughly 40 million units in January to about 100 million by May, projecting 600,000 to 800,000 units shipped by 2028.

The bandwidth claim looks smaller once you widen the frame. A German research institute demonstrated a waveguide-coupled germanium photodiode running at 265 GHz back in December 2021, noting commercial germanium photodiodes were already running around 50 GHz - both well past the "greater than 40 GHz" this filing offers as an advance over a 12 GHz baseline. A separate, more recent design reaches the same territory a different way, splitting light across two waveguide inputs instead of angling one facet, at about 58 GHz. None of that makes the angled facet wrong, but it does mean, to my eye, that the number Ranovus chose to publish was the flattering one.

What would have to be true

A few things would have to hold up that I haven't seen resolved anywhere in what's public. Integrated lasers would have to prove reliable enough to survive next to hot switching silicon at the rates hyperscalers actually run their gear, despite data suggesting otherwise. The co-packaged optics market would have to grow toward the higher end of estimates researchers can't agree on. And a photodetector built for 40 GHz would need a place in a market where faster ones already exist in the published literature.

My best guess: the patent is real, useful engineering, and also, separately, a subsidized company doing what its funding expects of it - neither fact cancels the other out. Betting the whole strategy on the laser staying onboard doesn't strike me as obvious, but I don't run a photonics fab, and the people who do are the ones who will find out.

Sources

Reporting for this piece drew on the published filing and the references below.

1. RANOVUS delivers industry's first 6.4Tbps Co-Packaged Optics with integrated laser
2. Why Co-Packaged Optics Uses External Lasers Instead of Integrated Sources
3. Ranovus Expands Ottawa, Ontario, Semiconductor Operations
4. Canadian government ups semiconductor focus, invests $36 million in Ranovus
5. Germanium-on-Silicon Waveguide-Integrated Photodiode with Dual Optical Inputs
6. Co-Packaged Optics Is Not Delayed. SemiAnalysis Is Just Wrong
7. Co-Packaged Optics Market Size & Share Report
8. Co-Packaged Optics Market Report
9. World record: Ultrafast germanium photodiodes

The record, briefly. US20260259369A1, “Waveguide Photodetectors For Reducing Retro-Reflection And Increasing Bandwidth” Discloses a waveguide photodetector with a photodiode combining a main waveguide and an absorbing waveguide flanked by n-doped and p-doped semiconductor regions, featuring an input facet angled to refract incoming light toward the p-doped region, reducing retro-reflection and increasing bandwidth.

More angles on this

• Optical Processing System Questions the Default ArrangementAnother reading from the same G02B classification family, optical elements, systems or apparatus.
• Luminous Laminated Glass Panel Comprising A Functional Coating Questions the Default ArrangementAnother reading from the same G02B classification family, optical elements, systems or apparatus.
• CPC G02B: Optical elements, systems or apparatusThe official classification family this record was filed into, published in the USPTO Cooperative Patent Classification scheme.
• CPC H10F: Inorganic semiconductor devices sensitive to infrared radiationThe official classification family this record was filed into, published in the USPTO Cooperative Patent Classification scheme.
• Inside multi-tiered semiconductor waveguideAnother reading from the neighbouring G02F classification family, optical devices or arrangements for the control of light by modification of the optical properties of the media of the.

Read around this one:Public patent recordUSPTO published figure