Towards Silicon Photonics 4.0

Last week I had the privilege to present a plenary talk at the 50th edition of the European Conference on Optical Communication (ECOC) in Frankfurt, Germany. The title of the talk was ‘Towards Silicon Photonics 4.0’ (#SiPh4.0).? Here I summarize some of the key ideas presented in this talk.

Silicon photonics has become a key enabler for high speed pluggable optical transceivers for data-communication and telecommunication, with aggregate data rates flirting with the 1Tb/s level. Millions of such transceivers are being manufactured each year by tier-1 foundries and IDMs, and find their way (predominantly) to data centers. The dominant technology is based on silicon-on-insulator (SOI) wafers (200 or 300mm) that are processed in a CMOS fab with a 45, 65 or 90 nm node. In parallel there is a large momentum of activity in R&D on new functionalities, applications and markets outside the communications area. This includes high performance computing systems and AI, industrial and medical sensors, diagnostics, LIDAR and 3D sensing, quantum applications, and more. In many cases these new functionalities cannot be implemented on the basis of the standard SOI-technology but require a somewhat different or more advanced process flow, often with new materials so as to enable new functions or better performance. The fact that many applications operate at spectral bands other than the telecom O- and C-band is a major driver for the introduction of new materials. The list of new materials is long but includes silicon nitride, aluminum oxide, aluminum nitride, lithium niobate, barium tantalate, graphene and other 2D materials, polymers, liquid crystals, colloidal quantum dots, and much more. This is all good and well in the research arena, but the development of an industrial supply and value chain for this large diversity of materials is a gigantic challenge, especially considering that some of these materials are not so welcome in a CMOS-fab. This explains in part the big gap in traction between research and product development for new applications on one hand and actual product sales and new industrial process flows on the other, as illustrated in the graphic. If there are a hundred new product types being researched and developed (outside the transceiver area), there are probably less than ten that are being sold in the marketplace today. Among many reasons for this gap, I believe that one stands out: the major barriers that fabless startups are facing when developing a product based on a still immature industrial supply chain. Often this implies that part of the NRE budget of the startup needs to be spent on co-investment in a new process flow by a technology provider, which can easily be too expensive for a startup.

This sounds like a tough challenge. At SPIE Photonics West earlier this year I already discussed the problem, but did not get very far in suggesting solutions… At ECOC I tried to articulate possible ideas towards a solution. I named it ‘Silicon Photonics 4.0’ (#SiPh4.0).?

In essence #SiPh4.0 pertains to any measure that breaks down supply chain barriers, especially for innovative start-ups and scale-ups, in spite of diversifying technologies, applications and markets. The term is obviously inspired by the notion of Industry 4.0, which in essence boils down to smart manufacturing, with a mix of dimensions, including agile and decentralized supply chains, extensive use of digital technologies, automation and AI, additive manufacturing, workforce upskilling, and more. Translating this to silicon photonics, one can envisage a series of approaches, that create the best possible conditions for an economically sustainable supply and value chain of a diverse silicon photonics ecosystem. The list of approaches, listed below, is just my personal first attempt for this matter. I am sure it will evolve over time, and I welcome feedback and contributions from the whole community to refine and extend it.

Here is my list:

• a decentralized open access foundry and supply chain model
• additive chiplet-based backend manufacturing approaches
• technology service providers that offer robust technology modules, with the following attributes: standardized modules that can be reused for multiple applications; data driven, thorough, open-access module-PDKs; elaborate specification of “interfaces” between modules; parametrized models, for designers to use seamless interfacing to non-photonic modules (electronic, mechanical, chemical, fluidic…); process control monitor circuits (PCMs) to assess compliance of individual modules with specs
• development of multi-stakeholder PDKs for complete process flows
• programmable PICs for rapid prototyping
• upskilling and reskilling of personnel

One may argue that there is nothing really new in this list, but I do believe that individuals actors in the supply chain – both providers and users – can make choices that strengthen, more than other choices, the desperately needed route towards order-of-magnitude growth in this field. Underlying such choices there will always be the dilemma between short-term and long-term optimization of revenue.

The above mostly focused on the chip manufacturing. Needless to say that similar challenges exist in chip packaging, assembly and test. So, it makes sense to broaden SiPh4.0 to all technologies needed to design, build and test products based on silicon photonics chips.

Once more, I welcome feedback to the ideas presented here, as well as additional ideas. Please use the hashtag #SiPh4.0

Let's PIC it, as they say.