A second source is a new part.

A second foundry, on paper

A module maker ships a silicon photonics transceiver built on one foundry’s platform. Orders double. The supply manager wants a second foundry for the same photonic die, and a second assembly house to package it. In the purchasing plan, this is two new contracts and the same part number.

JESD264 does not see the same part. The standard is the first industry-wide qualification standard for silicon photonics, and JEDEC published it on 1 October 2026. Its first clause says when a qualification must happen again. Under that clause, both new suppliers start a new qualification.

This essay follows that clause. A previous essay read the test menu and the statistics of its sample plans. This one reads the scope: who must comply, which events restart the obligation, and what the standard leaves out. The source is a close reading of clause 1 that Latitude Design Automation, a photonic design-software company, published on 19 September 2026. Latitude worked from the JC-14.3 committee document of August 2026, before the JEDEC announcement of 1 October. We did not read the standard itself, so every clause 1 statement below is that reading, not a quotation.

Who the standard binds

Integrated circuits divide reliability work in a fixed way. The foundry qualifies its process and publishes a process design kit (PDK), the set of rules and verified components that designers use. A customer who designs inside those rules inherits the foundry’s reliability work.

Silicon photonics breaks that division. The reading makes three points:

• Most silicon photonics customers do not design from a mature, foundry-defined PDK. Bend radii, grating coupler shapes and modulator doping profiles are often the customer’s own intellectual property.
• Some foundries now offer qualified photonic PDKs. JESD264 says plainly that these are not yet as mature as electrical PDKs.
• So nobody can assume that the foundry owns the reliability of a component or a layout. Without a clear agreement, the reading concludes, that responsibility moves to whoever designed the optical circuit. That is often the end user, not the foundry.

The standard therefore names its parties broadly. It binds foundries that make silicon photonics wafers and the OSATs (outsourced assembly and test houses) that package them. It also binds end users. That includes die integrators such as datacom and telecom module makers. It also includes any company that assembles a multi-chip module around a silicon photonics die.

Four events that restart the clock

Clause 1 lists the events that require qualification. A new product is the obvious case. The reading lists four more:

1. A new silicon photonics supplier enters the supply chain.
2. A qualified supplier moves production to a new site.
3. A qualified supplier changes its production equipment.
4. A new non-standard design appears. The standard’s example is a discrete component that changes from one technology to another.

Put these events next to the parts of a supply chain, and a pattern appears. Each element of the chain has its own trigger:

The consequence is the central finding of this essay. A JESD264 result does not belong to a design. It belongs to a configuration: one design, from one wafer supplier, at one site, on one set of equipment, through one assembly house. Change any one of the five, and the qualification starts again. The reading gives a concrete case. A component passed at foundry A two years ago. Then foundry A moves the line, changes a tool, or sends the same design to a second OSAT. Nobody can assume that the component is still qualified.

What a restart costs

How much testing each trigger repeats is not in clause 1. The opening clauses set no test conditions or sample sizes. Those come later in the standard, after it defines a mission profile, the use conditions that the tests stand for. A small process change may need only part of the plan. So the arithmetic below is an upper bound, not a quotation.

The draft that Cisco presented in July 2026 gives the full plan for a silicon photonics die or chipset sold to an end user:

• Three lots.
• 77 parts per lot for high temperature operating life (HTOL), a test that runs parts powered and hot to accelerate wear.
• 25 parts per lot for the other stress legs.
• A pass/fail point for HTOL at 2000 hours.

One full run is therefore 306 parts. The HTOL leg alone is 2000 hours of continuous stress, about twelve weeks, before a single result exists. Lot builds, test setup and analysis come on top of that.

Now go back to the supply manager. Each pair of foundry and assembly house that ships the part is one configuration, and each configuration needs its own qualification. The count depends on which assembly house each die goes to:

• Two separate chains, each foundry with its own assembly house: two qualifications.
• Either die through either assembly house: four qualifications, and at the full plan, 1,224 parts.

The calendar matters more than the part count. Each new source waits at least one HTOL cycle before its first qualified shipment.

Qualification work is not new. OpenLight’s published GR-468 results on Tower’s platform show what one run involves. What changes is that the restart now has a public definition. A buyer can point to clause 1 and ask a supplier which configuration its result covers.

Two kinds of component, two kinds of failure

The reading also explains why one test flow cannot cover every component. The standard divides silicon photonics components into two classes:

• Passive components carry light and do not modulate or detect it. Examples are waveguides, splitters, couplers, polarization splitter-rotators, and multiplexers and demultiplexers. The failure to watch is insertion or propagation loss that drifts with time and temperature.
• Active components convert between electricity and light, or use an electrical bias to control light. Examples include integrated lasers, modulators, photodiodes and thermal tuners. They inherit failures from both domains: leakage and dark current on the electrical side, and phase drift and loss of responsivity on the optical side.

So JESD264 does not write one test flow for “silicon photonics components”. Later clauses give each component type its own qualification table.

What a JESD264 claim does not tell you

Clause 1 is as clear about the boundary as about the scope. Per the reading, the standard does not cover:

• Harsher use conditions: military, automotive and avionics. A design for those environments needs more or different qualification.
• RoHS and lead-free compliance.
• Production test, measurement and inspection.
• Production burn-in and production temperature cycling.
• Regulatory compliance and safety.
• The qualification of piece-part optical devices on their own.

The last exclusion needs care. JESD264 qualifies the assembled silicon photonics die or chipset. That includes the parts that make it work: lasers attached by eutectic or hybrid bonding, lenses, isolators, and integrated circuits. The qualification covers a lens or an isolator as part of that assembly, not alone. So a JESD264 result on a chipset does not qualify the lens supplier for another product.

The production exclusions matter for a buyer too. JESD264 is a qualification standard. It shows that a configuration can survive its stress tests. It does not say how a supplier screens each shipped unit. That is a separate question for every supplier.

What to take away

1. Under JESD264, a qualification belongs to a configuration: design, wafer supplier, site, equipment and assembly house. Each element has a trigger.
2. A second foundry or a second assembly house is a new qualification. A full repeat of the draft plan is 306 parts and about twelve weeks of HTOL.
3. Where no contract assigns reliability, it falls by default on the designer of the optical circuit. Put the assignment in the contract.
4. Ask which configuration a result covers, and when its last trigger happened. A result from before a site move or a tool change does not cover the current configuration.
5. A JESD264 claim does not cover production screening, harsh environments, or loose optical parts. Ask for those separately.

Multi-sourcing is how the optics industry protects supply. JESD264 does not stop it. It makes the cost visible and puts a date on it. In silicon photonics, a second source is a new part.