🏒Data Centers
News Brief
PIC100 TSV platform
optics integration
near-packaged optics
co-packaged optics

PIC100 TSV Platform: What Near- and Co-Packaged Optics Mean for Infrastructure

InfraSale Editorial
March 10, 2026
52 views
Data Center Dynamics

Discover how the PIC100 TSV platform will revolutionize optics integration in technology!

The bottleneck in high-performance computing has never really been the chip; it's been getting data *off* the chip fast enough to matter. As AI workloads push switch and server bandwidth requirements into the terabit-per-second range, the interconnect β€” the physical pathway between silicon and fiber β€” has become the defining constraint. The PIC100 TSV platform is a direct answer to that constraint.

Here's what makes it worth paying attention to: this isn't an incremental spec bump. Through-Silicon Via (TSV) technology combined with photonic integrated circuit design represents a fundamental rethink of how optical engines get built, packaged, and integrated into the systems that run modern data centers.


What the PIC100 TSV Platform Actually Is

TSV stands for Through-Silicon Via β€” a vertical electrical connection that passes directly through a silicon die rather than routing around its edges. In the context of photonic integrated circuits, this matters enormously. Traditional packaging routes electrical signals laterally across a substrate before they reach optical components, introducing resistance, inductance, and signal degradation at every turn.

TSV architecture shortens those paths to near-zero, which means lower power consumption, higher signal fidelity, and the ability to stack functional layers in ways that were previously impractical.

The PIC100 platform is built around this principle from the ground up. Rather than adapting existing photonic packaging to accommodate TSVs as an afterthought, the architecture is designed to support two distinct but related integration paradigms: near-packaged optics (NPO) and co-packaged optics (CPO). Each serves a different point on the performance-versus-practicality curve.


Near-Packaged Optics vs. Co-Packaged Optics: The Real Distinction

These terms get used interchangeably in marketing materials, but they shouldn't be.

Near-Packaged Optics

Near-packaged optics places the optical engine in close physical proximity to the switching ASIC β€” typically on the same board, sometimes within the same module β€” but the optical and electronic components remain in separate packages. The electrical interface between them is short, measured in millimeters rather than centimeters, which dramatically reduces signal loss compared to conventional pluggable transceivers like QSFP-DD that sit at the faceplate.

The practical advantage here is manufacturing flexibility. Because the components are still physically separable, you preserve the ability to test, qualify, and replace the optical engine independently from the compute or switching silicon. For hyperscalers running systems at scale β€” where a single failed optical component can affect rack-level throughput β€” this is not a trivial concern.

NPO occupies a pragmatic middle ground. It captures most of the power and bandwidth benefits of tighter integration without forcing the industry to abandon existing supply chain and serviceability assumptions overnight.

Co-Packaged Optics

CPO goes further. Here, the optical engine and the switching ASIC share the same package substrate β€” sometimes mounted side-by-side in a multi-chip module, sometimes stacked using advanced interposer technology. The electrical interconnect between silicon photonics and the switching logic drops to the scale of a few hundred microns. At that distance, you can drive signals with far less power, and thermal management becomes a shared problem across the entire package.

The tradeoff is complexity. Co-packaged designs require optical and semiconductor components to be qualified together, which complicates testing and raises the stakes for any single component failure. Yield management across heterogeneous chiplets is genuinely hard.

What the PIC100 TSV architecture does β€” and this is the non-obvious part β€” is provide a common physical and electrical foundation that can serve both integration styles. Rather than designing separate platforms for NPO and CPO, a TSV-based PIC can be configured for either deployment, which meaningfully compresses development timelines and reduces qualification overhead for system integrators.


Why This Matters for Data Center Infrastructure

Scale the numbers, and the implications become concrete. A 51.2 Tb/s switch ASIC β€” the current leading edge, with 102.4 Tb/s already in development β€” requires hundreds of high-speed optical lanes running simultaneously. At that bandwidth density, conventional pluggable optics consume somewhere between 15 and 25 watts per port just for the transceiver. Multiply that across a 64-port switch, and you're looking at a kilowatt or more dedicated solely to the optical interface layer.

CPO implementations have demonstrated power reductions on the order of 30–50% per optical lane in research and early commercial deployments. At data center scale β€” tens of thousands of switches, millions of ports β€” those efficiency gains translate directly into reduced cooling infrastructure, lower PUE, and meaningful operating cost reductions.

For anyone building or financing data center infrastructure today, optics integration isn't a component-level decision β€” it's a facility-level one.

The transition also reshapes vendor relationships. When optics and ASICs are co-packaged, the clean separation between merchant silicon vendors (Broadcom, Marvell) and transceiver suppliers (Coherent, II-VI, Lumentum) starts to blur. System integrators and hyperscalers gain leverage to specify integrated solutions rather than assembling them from discrete components. That's a structural shift in how value gets distributed across the supply chain.


Where the Market Is Heading

The optics integration trend has been building for years, but the AI infrastructure buildout has compressed the timeline considerably. Hyperscalers that previously operated on three-to-five-year technology refresh cycles are now making architectural decisions annually. That urgency creates real commercial pull for platforms like PIC100 that can serve both near- and co-packaged deployments.

Analyst projections for the co-packaged optics market vary, but consensus estimates point toward significant volume ramp beginning around 2026, with CPO becoming the dominant form factor for high-radix switching by the end of the decade. Near-packaged optics will likely serve as the transition technology β€” capturing early adopters who need better performance than pluggables but aren't yet ready to commit to the qualification complexity of full CPO.

The TSV dimension adds another layer. As chiplet-based architectures become standard in both compute and networking silicon, the ability to vertically interconnect photonic layers through TSVs opens design space that simply didn't exist in planar packaging. Think optical engines that integrate directly with memory stacks or photonic switching fabrics that bypass the copper interconnect layer entirely.


What Industry Professionals Should Watch

If you're on the infrastructure investment or development side, a few signals are worth tracking closely.

First, watch hyperscaler procurement language. When RFPs start specifying CPO or NPO compatibility rather than transceiver form factors, the market has turned. That shift is closer than most people outside the industry realize.

Second, pay attention to which system integrators are building internal photonics expertise versus which ones are staying dependent on merchant solutions. The former group will have structural cost advantages as CPO volumes scale.

Third, the real bottleneck in CPO adoption isn't technical β€” it's ecosystem. Fiber management, repair protocols, and thermal specifications for co-packaged modules need industry standardization before volume deployment is practical. Platforms that contribute to or align with emerging standards (the Optical Internetworking Forum has active working groups here) will have shorter paths to customer qualification.

The PIC100 TSV platform's dual support for near- and co-packaged optics isn't just a feature list item. It's a recognition that the industry is mid-transition and that infrastructure built today needs to serve both where the market is and where it's going. That kind of architectural foresight tends to age well.


Explore the InfraSale Marketplace for cutting-edge solutions.


Related Topics:
optics integration
near-packaged optics
co-packaged optics

InfraSale Marketplace

Ready to act on this signal?

List a site or post a power requirement in under five minutes.