Arjun Kantimahanti
COO, Mixx Technologies
The silicon photonics processes running in production today were built for pluggable transceivers, whether four-channel or eight-channel designs, and the defect density targets that go with them. Co-packaged optics is a different endeavor, with high radix changing the arithmetic. A co-packaged engine carries 32 or 64 channels, making the distance between demonstrating a CPO product and deploying one a significant manufacturing challenge. We asked Arjun Kantimahanti what closing that distance actually requires.
His answer starts with two differences, the first being ownership. “In transceivers, the end customer owns the link budget,” he says. In co-packaged optics, it transfers to the company building the engine. The second is density, and the consequence of that follows: “Whatever is non-critical in transceivers becomes critical in CPO.”
These distinctions move past design. Co-packaged optics is new enough that the supplier is on the same learning curve as the product. The sequencing isn’t what the industry is accustomed to, Kantimahanti says. “CPO is not a solution where you can design the product, get the wafers or the chips from the contract manufacturers, qualify the process, and then start volume production,” Kantimahanti says. Run it that way, and because the technology is brand new, the supplier is not ready when volume production starts. Mixx starts from the opposite assumption.
Asked how Mixx approaches process qualification differently, Kantimahanti reframes the question. “Actually, it’s not process qualification,” he says. “It’s yield learning.”
Mixx works with its manufacturers from the outset, identifying and addressing yield issues during pre-qualification, rather than waiting for the ramp. “We have the product development cycle and the process improvement cycle, which is done at the supplier, and they both run in parallel.” Systematic issues surface early enough that the fab still has time to correct them. Mixx focuses on qualifying the product, while the fab matures the process that will carry it.
Kantimahanti joined Mixx after three decades building photonic processes, including senior roles at SilTerra Malaysia and Broadcom’s optical systems division. He says what drew him to Mixx was its answer to the hardest version of the problem: scale-up networks, where thermal requirements are stringent, and densities climb. “I was quite attracted to the Mixx 3D approach.”
Where much of the industry grows the reticle outward to win beachfront density, Mixx stacks vertically at wafer scale, shortening the signal path rather than widening it. Integrating at once also removes the epoxies and underfills that 2.5D depends on, materials that carry a high risk of reliability failure under scale-up thermal conditions. Removing them, in Kantimahanti’s words, “actually makes reliability better.” And it changes the unit of measure: optical assembly is conventionally counted in hours per unit; Mixx is changing that to units per hour, bringing a semiconductor metric into optics.
“Over the next year, the main challenge for us is to demonstrate this 3D integration,” he says, “where we plan to put all of this together at the wafer level.”
