As the PIC Ecosystem Grows, So Grows the Future Prospects for PICs

One debate that continues within the integrated photonics community is how far, or rather how soon, integrated photonics will transform data centres into more efficient, monolithic computing platforms for AI.  

To be clear, integrated photonics has long formed the interconnect between data centre racks and switches. But “integrated” has come to mean something different as Photonic Integrated Circuits (PICs) have shifted optical components and interlinks onto single semiconductor dies. If advanced far enough, such integration could help usher in massive data centres built on light-speed compute. 

Industry/academic consortia and startups are no longer the only voices evangelising the potential of PICs. Semiconductor bellwethers such as TSMC, NVIDIA, and Broadcom have all invested in initiatives to expand the technology, and even hyperscalers like Google, Microsoft, Amazon, and Meta are sharing how photonic functions are penetrating deeper into their networks. 

The expanding range of players exploring PIC technology stands as a stronger testimony to the promise of PICs than the work of any single company, no matter how large. Why? Because, as with the semiconductor industry, the complexity involved in producing and implementing PIC technology demands support from a similarly complex ecosystem. The challenges to design, manufacturing, packaging, and implementation are legion throughout the value chain.  

Design by collaboration 

Photons and electrons interact very differently. So do the engineers who work with them. Designing systems around photons and electrons requires very different skill sets, materials, and even languages. Traditional semiconductor designers, tasked early on with integrating photonic functions into their chips, suddenly had to contend with unfamiliar ideas like crosstalk, dispersion, and nonlinear effects. That “culture gap” is diminishing, however, as new software platforms have become more able to accurately model the behaviour of light within tiny circuits.  

However, the application-specific nature of PICs still requires designers to collaborate closely with downstream partners, such as chip fabs, packaging foundries, and even end users, to anticipate potential problems early in the design process. The industry will benefit from more automated and standardised software design tools. Software vendors such as Synopsys, Luceda Photonics, and GDS Factory are helping to address these challenges with seamless design flows that extend from PIC tape-out to manufacturable designs. 

Standardised manufacturing 

Traditional CMOS chip fabs arguably confronted an even wider culture gap when trying to incorporate photonic components into their production processes. As previous posts have pointed out, integrating photonic functions onto familiar silicon substrates requires the use of materials still unfamiliar to the fab. Indium phosphide (InP), for example, forms the lasers powering telecommunications networks as well as most PICs. Thin-film lithium niobate (TFLN) or barium titanite (BTO) enables ultra-high-speed on-chip modulators, and silicon nitride (SiN) offers low-loss waveguides that carry optical signals across these chips. 

Traditional chip foundries, such as GlobalFoundries, Tower Semiconductor, Intel, and NVIDIA, have developed or deployed various hybrid integration schemes to incorporate photonic components based on these materials. But all introduce issues related to yield and manufacturing throughput, which, in turn, diminish the already razor-thin margins of high-volume chip production. 

As with design, standardising CMOS processes that leverage these new material technologies will be key to smoothing their adoption for manufacturing. Individual PIC fabs already offer Process Design Kits (PDKs) and pricing schedules for wafer runs. But industry-wide standardisation will help further. 

Meanwhile, an emerging vanguard of companies is advancing InP, SiN, TFLN, and BTO platforms as alternatives to silicon wafers. Both Coherent and PhotonDelta partner, SMART Photonics, for example, are developing InP wafers of increasing diameter to make them more cost-competitive with silicon. 

New Origin and LioniX International— also part of the PhotonDelta ecosystem — are pioneering SiN platforms for PICs. And, yet another PhotonDelta partner, Rapid Photonics, is leveraging TFLN and BTO to develop PICs embedding ultra-high-speed modulators. 

Testing needs automation 

Often overlooked in discussions about the challenges of PIC manufacturing, the ability to test bare dies, wafers, and packaged devices is both critical and difficult to perform efficiently. As with semiconductor chips, early functional testing is important to avoid the cost of finishing and packaging bad chips. But unlike semiconductor chips, PICs require mixed-signal electro-optical characterisation. Standardisation is again important, but so is automation, which is the only cost-effective way to maintain speed and accuracy when testing optical alignment, insertion losses, and high-speed modulation. Companies like EXFO, and MPI Corporation are developing faster and more reliable testing techniques to address these issues. 

Packaging begins early 

The transformative functionality of a PIC relies as much on its packaging as it does on the die within. But compared to conventional IC enclosures, PIC packaging is made more complex by the need to accommodate multiple material platforms and both electrical and optical interfaces, as well as the time-consuming alignment and coupling of optical connections. Thermal management is another significant challenge due to the high power density of PICs. It is no wonder that PIC packaging contributes the lion’s share of overall manufacturing costs. 

The challenges are substantial enough that GlobalFoundries announced that it would dedicate $575 million to build an Advanced Packaging and Photonics Center to perform advanced assembly and testing functions in addition to packaging. 

PHIX, a PIC packaging foundry, contract manufacturer, and PhotonDelta partner, is also helping PIC developers optimise their packaging early with the introduction of open architecture prototyping packages (aka characterisation packages). The idea is to provide off-the-shelf building blocks and standardised production processes to let developers characterise early designs of chips or test for system integration.  

Another PhotonDelta partner, Epiphany Design, will introduce a different twist on the building block approach. The company developed a platform-agnostic hybrid laser module that can readily be integrated into PIC designs. By eliminating the need to design the laser component from scratch, Epiphany’s standardised module reduces testing, validation and overall development time. 

MicroAlign is yet another PhotonDelta partner helping to reduce PIC packaging design cycle with a solution that helps speed the alignment of PIC packaging optical interfaces by bringing individual fibre position control to a fibre array. 

It takes an ecosystem 

The supply chain for PIC components might not be nearly as mature as that for electronic components, though it is light years ahead of where it was just a decade ago. Securing high-purity materials, specialised fabrication equipment, and even skilled labour will continue to be both a challenge and a metric for progress.  

It’s PhotonDelta’s mission to build an ecosystem for Photonic Chip technology to provide global players with an end-to-end value chain encompassing materials, design, manufacturing, test and packaging solutions. Moreover, we foster the collaborations that can enhance the development and value of new PIC implementations. Reach out to us to explore synergies and become part of the next big thing. Go to www.photondelta.com/contact/