Loading...
Technical Insight

Magazine Feature
This article was originally featured in the edition:
Issue 5 2026

End-to-end process flow under one roof

News

Turning a photonic chip into a product means bringing together photonics, electronics, packaging and test. Handoffs between separate actors are where processes often stall. End-to-end process flow control under one roof may offer the best way through.

by Antonio Teixeira, Co-founder and CSTO; PICadvanced

For simplicity, let’s consider a company that sees an opportunity and wants to see it developed, and this opportunity has in its core a PIC. The product can be, for example, a photonic integrated circuit that will move data faster, bring higher accuracy to sensing applications, or shrink into a single chip an optical system that today fills ten times the space. The process, as always, starts with a block diagram and some simulations, followed by a design that makes sense and is feasible in the chosen platform. On paper, the hard part looks finished: the physics holds up, the simulations agree, the layout is clean. However, when we start going forward, more and more difficulties start to pop up and quickly, it is understood that the chip design and concept were the easy part.

Building a photonic product is not the same as building a photonic chip, because a PIC will usually require electronic control and, eventually, opticalI/Os. It needs electronics to drive its components and read its monitors, a packaging structure that holds optical and electrical connections in micron-accurate alignment, a thermal design that stops components from drifting, and a system or subsystem test regime that proves that the whole assembly behaves– not once at room temperature, but repeatably, across the product’s working life, standards and client requirements. Each of those steps is a demanding discipline in its own right, and each, in today’s industry, due to the required investments and specialization tends to sit inside a different company.


Figure 1: The fragmented journey – crossing several separate specialization domains, where interface mismatches, IP boundaries, minimum order quantities, re-qualification and schedule slip are part of the stalling mechanisms.

The handoff chain
The journey from idea to product runs as a relay race between specialists who rarely share a building or a schedule (see Figure 1). The PIC itself is fabricated by a foundry, on dedicated runs or multi-project wafer cycles, with fixed design rules, with variable waiting times and queues. The driving electronics come from a separate design house, which must understand the requirements while working with its own tools, roadmap and timeline. Typically, the assembly and packaging stages are also outsourced – sometimes to an outsourced semiconductor assembly and test provider, or OSAT – with specific tooling and minimum order quantities written for high-volume parts, and the more tailored and medium and low-volume packaging houses, besides being expensive, are not fully available and also have their own timelines. Testing and qualification may involve yet another party and may require some feedback to the packaging entity in order to improve and validate the final requirements. Only then does a system integrator try to make the result fit the application it was meant for with high-level system testing and comparison with the required results.

Although this approach is feasible and potentially valuable because each party specialises in its own domain, efficiency can be lost in the transitions, simultaneously in time alignment and requirement completion. Also, intellectual property, process control and process feedback for continuous improvement can be a huge barrier to a smooth and seamless flow. What has been seen as a simple project becomes very complex and demanding for the company specifying the product, as well as for the companies that are developing each of the processes, since they have to fully understand the requirements that are passed from different levels of complexity and also different languages and environments. Individually, none of these is fatal; collectively, they open the gap between a chip that works and a product that ships.

The backdrop of these interdependencies is that this obstacle course is particularly demanding for those trying to push boundaries in photonic integration. The organisations with the sharpest application insight (start-ups chasing a new architecture, system vendors with a specific performance target, network operators who know exactly what their traffic demands, research groups spinning out a breakthrough) are usually the ones least equipped to run this complex full cycle. They have the idea, but not a multidisciplinary team spanning photonics, electronics, packaging and test, nor the volumes that persuade a large-scale supplier to adhere to the journey of changing or adding new processes, which is a task required for most of these more specific developments. The established transceiver vendors, understandably, optimise for their own high-volume catalogues, and a customised design at low or medium volume is an awkward fit for a business built on scale. That leaves the innovator to become a project manager across multiple suppliers who do not talk to one another, and none of whom fully grasps the idea behind the product.

Figure 2: Seamless multidisciplinary co-development under one roof – the same disciplines run concurrently around a single feedback loop, with the partner acting as accelerator between the outsourced foundry and the OSAT.

End-to-end process flow control and internal co-development, not coordination
The topic that we are addressing is already being adopted by several actors of each of the steps, especially by adding functions in the neighbourhoods. Packaging already incorporates some electronics development, as well as testing. However, these processes may be taken even further. It may start by refusing to treat design, electronics, packaging and test as a chain of separate contracts. Instead, they operate within a single internal co-development process: what we call working under one roof.

The distinction matters, because the real value lies in co-design throughout the different steps of the process. When the packaging engineer sits beside the photonic designer, package constraints shape the chip layout from the first day rather than the last, commonly defined as design for packaging (DFP). When the test engineer is in the room while the architecture is still being developed, the device is born testable instead of being forced onto a probe station months later, design for testing (DFT). When the electronics designer knows all implications stemming from the PICs and the surrounding electronics and packaging, e.g. the package’s parasitic budget from the outset, the electronic control circuits are matched to the system or subsystem rather than compensating for it later. When the thermal modelling and the electronics design inform each other continuously, parameter drift is engineered out early rather than discovered late. Problems that would otherwise surface at a handoff, expensive and slow to fix, or even requiring a recall and redo of the full or partial flow, are caught while they are still cheap and, therefore, more suitable to be in time for the market window.

In practice, this means treating manufacturability and testability as architectural questions rather than downstream constraints. How a device will be packaged, aligned and measured shapes the design from the first day, and settling those questions alongside the PIC itself is what keeps them from becoming late-stage problems.

This is the model we have built our business around. Rather than acting as one more link in the relay, we position ourselves as an accelerator between the foundries and the OSATs (Figure 2): fabless, so the wafers are still made by the expensive and specialised fabs, and deliberately not chasing commodity volume, which the OSATs handle at scale. What we own is everything in between – architecture, electronics, advanced packaging, thermal management and test carried from concept through to a qualified pre-series. In other words, we take ownership of exactly the integration gap that used to swallow projects whole.

“Under one roof” does not mean that every manufacturing step must happen in one building. It means that one integrated engineering team owns the interfaces, the trade-offs and the outcome. Foundries and volume manufacturing partners remain essential parts of the ecosystem; the difference is that the customer no longer has to manage the engineering gaps between them.

Consider what this means in practice for an access-network transceiver, the kind of module that carries broadband into homes (Figure 3). Getting one to market the traditional way means aligning a photonic chip, custom driver and receiver electronics, a fibre-coupled package and a full qualification programme, each from a different supplier, each on a different timeline. Handled as a single co-development, the same job becomes one iterative conversation: the package is designed around the chip, the electronics around the package, the test plan around all three, with one team adjusting each as the others evolve. The result is a faster and smoother path to a pre-series that a real network can actually deploy, and the same logic scales up to the dense, high-speed interconnects that AI datacentres are now demanding.

Figure 3: From chip to product: a photonic-based optical transceiver for access networks.

The feedback loop
For us, co-development is not the same as coordination. A prime contractor who simply subcontracts the same fragmented chain still inherits every seam in it; the customer gains a single point of contact but none of the underlying integration. True co-development collapses the loop because design, foundry liaison, packaging and test share one team and one feedback loop. For example, a problem found in test on a Monday can reshape the layout by Friday, instead of triggering a three-month round trip across three companies, each waiting on the others and synchronizing their own roadmaps. Tighter loops mean fewer and faster iterations; fewer iterations mean lower cost and earlier time to market. And, above all, someone owns the outcome from end to end. The economics matter as much as the engineering. A late interface problem can trigger another fabrication cycle, a package redesign, new tooling or a fresh qualification round. Preventing one such iteration can be worth far more than optimising the cost of any individual component. The relevant metric is therefore not simply engineering cost, but the cost and time required to reach a manufacturable product.

There is a human dimension to this as well, because a chain of arm’s-length transactions loses a little information at every boundary, and a little trust with it. A co-development relationship works the other way: the customer’s deep knowledge of the application and the partner’s deep knowledge of integration reinforce each other, pulling in the same direction rather than negotiating across a contract. The customer keeps their intellectual property and their differentiation (the reason their product deserves to exist), while the partner supplies the plumbing that turns that differentiation into something manufacturable. Neither has to become the other to get a product out of the door.

None of this works without real capability behind it. Our cleanroom and automated optical and electrical test stations (Figure 4), and the advanced packaging platforms that sit alongside them, are at the service of this model, not the model itself. The model is what turns a decade of assembled expertise into a way for someone else’s idea to reach the market.


Figure 4: Inside the pipeline: advanced packaging and testing in the PICadvanced cleanroom.

The fair objections
Two objections deserve a straight answer. The first is dependency. A chain of separate suppliers gives a customer leverage and a second source, while a single integration partner concentrates that risk in one relationship. The trade-off is real, and it is why the deliverable of a co-development should be a transferable package - a documented architecture, a qualified process and a test strategy that another party can pick up - rather than a black box only one company can reproduce. Nor does co-development imply keeping production under one roof indefinitely: the objective is to remove integration risk before scale, and once the architecture, package, test strategy and manufacturing process are stable, production can move to the manufacturing partner best suited to the volumes required. Integration risk is worth removing; lock-in is not worth acquiring.

The second is best-of-breed: a specialist at each step may well beat a generalist at that step. Often true, and eventually irrelevant if the parts do not meet. The argument for co-development is not that any single discipline is performed better under one roof, but that the interfaces between them are owned by someone. Where a design is conventional, the interfaces are standard and the volumes are high, best-of-breed wins on merit, and a commodity module built to a settled specification is better served by the scale economics of a large transceiver vendor. Where the interfaces are the hard part - a genuinely new design, low to medium volumes, risk concentrated in the seams – ownership of those seams beats excellence at the ends. That is precisely where most photonic products live today.

The bottleneck is shifting
As photonic integration spreads, pushed by AI-driven bandwidth growth, co-packaged optics, optical sensing and the first stirrings of quantum photonics, the bottleneck in the field is moving. As foundry access, mature process platforms and process design kits make PIC development increasingly accessible, a growing share of product risk is migrating downstream, from proving the photonic function to integrating it into a reliable, testable, thermally stable and manufacturable system.

And manufacturable ultimately means manufacturable at acceptable yield. A design that performs brilliantly once but cannot be assembled and tested repeatably is not yet a product. Yield is therefore not simply a manufacturing metric; it is an architectural constraint that needs to enter the co-design loop early.

In the next decade, the organisations that thrive will not necessarily be those that can fabricate the cleverest chip, but those that can co-develop across every discipline that stands between that chip and a customer. That points towards a genuine democratisation of photonics.

The role of EDA tools is also a relevant matter, since they are the ones able to incorporate most of the processes in one single tool and therefore able to pass requirements in a standardized manner to the relevant actors.

A founder with a good idea and no cleanroom, or an operator with a precise need and no photonics team, should be able to bring a PIC-based product to market by plugging into a partner who carries the integration burden for them. Building a photonic product is still hard – the physics, the electronics, the packaging, the heat and the testing all remain hard-won. The difference is that the effort no longer has to be scattered across a dozen companies and several regions. It can live under one roof. And for the engineer with the genuinely good idea, that changes everything: the chip can go back to being the interesting part.


Logo
x