Photonic IC sourcing
Silicon Photonics vs InP: Choosing a Foundry Route for a Constrained PIC Process Step
The public record does not rank silicon photonics against indium phosphide; it supports an access-path comparison. Fit turns on the purchased artifact, the PDK or MPW entry path, evidence of operating versus planned capacity, and a named packaging acceptor.

The decision in brief
Neither route wins on the public pages, because no supplied source benchmarks silicon photonics against indium phosphide on performance, cost or yield. What the sources do establish is an access-path comparison: a silicon photonics foundry route fits the wafer-fabrication portion when the buyer needs a 200mm silicon photonics platform and can plan around announced 300mm scaling, while an indium phosphide pure-play route fits when the buyer needs epitaxial growth and processing and can plan around capacity expected in 2028.
Neither page names the packaging acceptor, so a separate module or test, assembly and packaging path must be identified before an RFQ. That decision rule is an Atom proposal, not a supplier conclusion.
What the silicon photonics foundry route supplies
GlobalFoundries announced on a November 17, 2025 dateline, in a release published November 18, 2025, the acquisition of Advanced Micro Foundry, a silicon photonics foundry based in Singapore. GlobalFoundries said the acquisition will expand its silicon photonics technology portfolio, production capacity and research and development in Singapore, complementing existing capabilities in the United States. The company said it will address long-haul optical communications, computing, LiDAR and sensing on AMF's 200mm platform in Singapore, with plans to scale to 300mm as market needs grow. The release also describes AMF as offering manufacturing, prototyping and testing services supported by proprietary technology platforms and process design kits designed in-house for telecom, data centers, LiDAR and sensors. GlobalFoundries further said it plans to establish a silicon photonics research and development center of excellence in Singapore with A*STAR, focused on next-generation materials for ultra-fast data transfer at 400Gbps. A buyer should treat this as a supplier announcement: it is not independent proof of available capacity or performance, and the 300mm scaling is stated as a plan tied to market needs.
What the indium phosphide pure-play route supplies
SMART Photonics describes itself as a pure-play manufacturer offering the complete production process of photonic integrated circuits, from epitaxial growth and re-growth through processing and polishing. Its published customer path has three steps: design with a process design kit, alpha production, then volume production. On capacity, SMART Photonics says it is investing in new 6-inch indium phosphide wafer production capacity, and that when fully activated in 2028, the new production is expected to manufacture more than 12,000 wafers per year, corresponding to millions of photonic chips. The page states this as an investment with an expected activation year, so a buyer should record it as planned capacity, not operating capacity, unless SMART provides a document showing otherwise. The page is undated, which is worth noting in the qualification file.
The access path before any reservation
AIM Photonics describes a gated sequence for purchasing wafer fabrication or test, assembly and packaging services, and the gates differ by customer status. For an AIM Photonics member, the existing membership agreement already covers the confidential disclosure that purchasing these services involves. For a non-member, AIM Photonics requires an executed non-disclosure agreement, signed by both parties, before anything else. Once that requirement is met, AIM Photonics asks for two more things: a services agreement setting the terms and conditions for payment and delivery of the hardware, and an export control form required under U.S. export control regulations. AIM Photonics states that a fully executed services agreement is required before it can reserve space on a multi-project wafer run or a private run, or provide test, assembly and packaging services. Agreements for AIM Photonics services are executed through the Research Foundation for the State University of New York. The inference a buyer should draw, labeled as such, is that an MPW listing or platform description is not immediately available capacity; access is a contractual and export-control process that has to be started early. Note that these gates are access conditions, not capacity evidence: the AIM page does not establish whether any run has operating or planned capacity behind it.
The packaging handoff
PHIX Photonics Assembly says it offers assembly services and contract manufacturing for photonic integrated circuits and MEMS, building optoelectronic modules based on indium phosphide, silicon photonics, silicon nitride and planar lightwave circuit platforms. PHIX states that a PIC by itself is not yet a functional product: packaging connects fibers, other PICs, cooling solutions and electronics in a micro-assembly process that yields a module with optical and electrical inputs and outputs. PHIX describes itself as the photonic equivalent of an outsourced semiconductor assembly and test company, with facilities in Enschede, the Netherlands and Bozeman, Montana, USA. The reasoned inference is that choosing a silicon photonics or indium phosphide wafer source does not select the packaging acceptor, and the PHIX page does not establish availability or qualification for any specific buyer. The buyer must name who accepts the package or module separately.
A proposed platform-fit memo
Atom proposes a one-page memo with five fields, completed per candidate before an RFQ. First, the purchased artifact: substrate, fabricated die, assembled module or test service. Second, the PDK or MPW entry path, including which agreements gate it and whether the buyer enters as a member or a non-member. Third, capacity status, recorded strictly as operating or planned, with the supporting document named; where a status is unknown, the field stays blank rather than being filled by an announcement or an agreement. Fourth, the packaging acceptor for the module. Fifth, open evidence gaps the candidate has not closed. This is a diligence method, not a technology ranking, and blanks should stay blank where the public pages do not establish a fact.
Evidence to request before an RFQ
Atom proposes that the buyer ask each candidate to identify the document that establishes whether quoted capacity is operating or planned, and to state that document's scope. An executed services agreement should be recorded as an access condition, not as proof of capacity status. Where a candidate offers volume by quarter, the buyer should record who commits to those figures and against what agreement. Publicly available artifacts, such as published process descriptions and any PDK documentation a supplier has already released without a confidentiality gate, should be collected before any confidential disclosure, so the requirement can be written precisely; the supplied pages do not establish that PDK documentation is generally nonconfidential, so the buyer should confirm what each supplier treats as public. Atom also proposes naming the acceptor for the package or module and attaching a boundary drawing before the RFQ is issued, so the fabrication-to-packaging handoff is unambiguous.
Conditions that would change the approach
The proposed rule shifts if the requirement changes. If the buyer needs epitaxial growth and re-growth as part of the process step, the SMART Photonics description points to the indium phosphide route. If the requirement is a 200mm silicon photonics platform with a stated scaling path, the GlobalFoundries announcement points to the silicon photonics route. If a candidate provides a document evidencing operating capacity, the planned-capacity caveat for that candidate drops. If the foundry itself accepts the packaged module through a test, assembly and packaging service, the separate handoff simplifies, but the AIM Photonics sequence shows that service still requires its own agreements. Atom Industries is development-stage and has no manufacturing capacity to offer today; this memo is research on how a documented requirement could match an experienced operator. The next useful question for the buyer: which document will each candidate sign to evidence its capacity status, and what does that document cover?
Sources & further reading
This perspective supports research and relationship development. It does not represent available manufacturing capacity, a committed transaction or an investment offering.