AanleidingNieuwe chipfabriek in Twente stap dichterbij
Based on J. Konstapel, “What Three Adds: Phase-Encoded Ternary Light in Symmetric Couplers and Counter-Coupled Resonator Chains” (Constable Research, Leiden, version 2.0, 21 October 2026): DOI: 10.13140/RG.2.2.36436.36483
A new chip factory is being built in Enschede. It is called New Origin. It will make photonic chips from silicon nitride. Chips that carry light instead of electricity. Imec is transferring its manufacturing process to the plant. The investment is around €200 million. Opening is planned for 2027.
The plant is a pure-play foundry. That means it makes other people’s designs. Any team with a good idea can have a chip made there. No cleanroom of their own is required.
This paper asks a focused question about that platform. Chips can encode data in the phase of light. Phase is the position of a wave in its cycle. You can divide a full cycle into any number of steps. The question is: what happens when you divide it into exactly three steps — and not two, four, or five?
The answer is a short list of exact, proven facts. Those facts are the raw material for new products.
2. The core idea, without the physics
A computer stores information in symbols. Today’s chips use two symbols: 0 and 1. This paper studies a third option: three symbols per unit, written into the phase of light.
Two results stand out for a general audience.
First: a simple optical device can read ternary data in one pass. A three-way beam splitter — a “tritter” — is a small, passive component. It can be built on a chip today. Feed it three light beams with ternary phases. The light itself does the sorting. Every valid pattern exits through one predictable port. Every invalid pattern spreads evenly over all ports. One device, one pass, no computation required. The paper proves this works only with three levels. With any other number of levels, the sorting breaks.
Second: simple networks of ternary elements have exact, predictable behaviour. Take two rings of ternary elements that read each other. They synchronise on chain lengths of 3, 9, 27, 81 — and on no other length. This is not a simulation result. It is a proven law, checked exhaustively on tens of millions of cases. Known behaviour means predictable products.
3. What you can practically build with it
3.1 Optical ternary logic and memory
The tritter acts as a hardware building block. It checks a ternary condition at the speed of light, without a processor. Possible uses:
- Pattern matching on ternary-encoded data streams.
- Arithmetic units for balanced ternary computers.
- Content-addressable memory read out in parallel.
Each use replaces many electronic steps with one passive optical step.
3.2 A new kind of problem-solving machine
Businesses already use optical “Ising machines”. These are specialised systems that solve hard optimisation problems — logistics, scheduling, chip design. They work with two phases, matching the binary Ising model.
This paper shows the ternary route. With three stable phases you get a Potts machine. That is the more general model. It handles problems with three or more natural states per variable. Think of scheduling with shifts, routing with priorities, or clustering with graded categories.
The paper adds a design warning that saves money. Such machines need millions of oscillators. On a chip, that does not fit. The arithmetic is simple. One million slow resonators need 750 to 15,000 cm² of silicon. A complete wafer offers 707 cm². The solution is a fibre loop: one loop of fibre, time-multiplexed, carries over 100,000 oscillators today and scales by length, not by area. This single insight redirects an entire design strategy.
3.3 Sensors and instruments
Three-phase encoding gives three distinct, robust reference states instead of two. That opens:
- Interferometric sensors with built-in error detection.
- Measurement readout that tolerates more signal loss before failing.
- Compact optical instruments for hospitals and labs, made affordable by the foundry model.
3.4 Demonstrator and education
A macro-scale tritter — mirrors, phase plates, a laser pointer — shows the sorting effect live on a desk. For a foundry, a university, or a trade fair, that is a powerful story.
4. The innovation, compared with current plans in Twente
The Twente ecosystem — New Origin, its backers, and the University of Twente — is currently about manufacturing capacity. Make more chips. Make them cheaper. Make them for others. That is a sound and necessary strategy.
This paper adds a product and architecture layer on top of that capacity.
| Twente’s current plans | What this paper adds |
|---|---|
| Manufacturing capacity for silicon nitride chips | A specific computing architecture that needs that platform |
| Open foundry access for external designers | Four concrete test designs any MPW customer can submit (Section 5.3 below) |
| Component-level quality: low loss, high stability | A proved mathematical layer: what the components can and cannot compute |
| Scaling by area: more chips per wafer | Scaling by time: fibre loops when area runs out |
| Binary optical computing (Ising machines) | The ternary, more general Potts-class route |
In short: the factory provides the road. This paper provides the destination and the map.
The competitive value is clear. Anyone can buy wafer capacity. Not everyone has a proved, tested design language for what to put on those wafers. A team that combines the two — exact ternary designs plus Enschede fabrication — holds something competitors cannot copy with money alone.
5. A concrete, low-risk first step
The paper proposes four structures that fit on one shared prototype die. That keeps the entry cost near the price of a single multi-project wafer slot.
- A tritter with three phase shifters. Tests the four-way sorting of all 27 ternary patterns. This is the go/no-go test for the whole approach.
- Three-level versus four-level encoding. Confirms that the even spreading — the key selling point — exists only at three levels.
- Ring chains of length 9 and length 10. Shows exact synchronisation at 9 and none at 10, as the lock law predicts.
- A ring pumped at 1231 nm. Tests conversion to the third harmonic, a step toward real three-phase oscillators.
Each test has a sharp predicted outcome. None is expensive. All can run through the foundry that opens in 2027. For an investor, this is a de-risked first milestone.
6. Honest limits
A good business case names its constraints. Three are stated openly in the paper.
- The lock law is exact for ideal elements with perfectly quantised phases. Physical resonators relax toward a state; they do not follow the ideal map exactly. Device test 3 addresses this gap.
- Optical oscillators with three stable phases have been proposed, and measured in superconducting microwave devices, but not yet demonstrated in a time-multiplexed optical loop. That demonstration is the main open engineering problem.
- Reading many windows in parallel through free-space optics needs a lenslet array, three diffraction orders, pixel-envelope correction, and three phase levels. It is possible, but it is a project, not a part.
None of these limits threatens the mathematics. They define the engineering roadmap.
7. Conclusion
The factory in Enschede gives the Netherlands a rare asset: industrial, open-access photonic fabrication. This paper shows what to do with it. It proves that the number three is not an arbitrary choice. Three-level light gives a passive one-pass reader, exact synchronisation laws, and a clean scaling strategy via fibre loops. It supplies ready-made test designs, a clear cost argument, and a stated set of open engineering problems.
The innovative value is not the wafer. It is knowing, with proof, what to put on it.
Annotated references
The occasion: the factory and the platform
- Van Dijk, B. (2026). “Nieuwe chipfabriek in Twente stap dichterbij.” Het Financieele Dagblad, 19 October 2026. The news item that anchors the paper: grid and water constraints resolved, €200 million investment. Why read: the market context for every opportunity described above.
- Imec (2026). “New Origin to collaborate with imec to scale photonic IC manufacturing in an industrial environment.” Press release, 13 March 2026. Confirms the LPCVD silicon nitride process transfer to Enschede. Why read: evidence that the platform will be industrially reliable, not just academic.
- LioniX International, “Silicon Nitride Waveguide Technology: TriPleX.” The waveguide family used in Twente; transparent from 405 to 2350 nm. Why read: the technical envelope of what designers can order.
- Liu, K. et al. (2022). “Ultralow 0.034 dB/m loss wafer-scale integrated photonics…” Optics Letters 47(7). Measured loss of 0.034 dB/m and quality factor 7.2 × 10⁸. Why read: the numbers that make long spirals and stable phases possible at all.
- Huang, G. et al. (2019). “Thermo-refractive noise in silicon nitride microresonators.” arXiv:1901.07112. The thermal drift numbers (19 pm/K, about seven times better than silicon). Why read: why stable phase encoding belongs on this material and not on silicon.
The application domain: optical computing
- Konstapel, J. (2025). “Left and Right Brain AI.” constable.blog. Introduces the oscillator-based computing layer on silicon nitride. Why read: the architecture this paper supplies the substrate facts for.
- Konstapel, J. (2025). “The Architecture of Right Brain AI (RAI).” constable.blog. Contains the 10⁶–10⁹ resonator specification that Section 4.1 of the paper recalculates. Why read: the origin of the one-chip assumption — and of its correction.
- Honjo, T. et al. (2021). “100,000-spin coherent Ising machine.” Science Advances 7. One 5 km fibre loop carrying 100,512 oscillators. Why read: the existing proof that time-multiplexing beats area limits — the recommended scaling route.
- Pierangeli, D., Marcucci, G. and Conti, C. (2019). “Large-scale photonic Ising machine by spatial light modulation.” Physical Review Letters 122. 40,000 phases on a spatial light modulator, read through one lens. Why read: the free-space alternative carrier.
- Honari-Latifpour, M. and Miri, M.-A. (2020). “Optical Potts machine through networks of three-photon down-conversion oscillators.” Nanophotonics 9. The proposal for three-stable-phase optical oscillators. Why read: the engineering route to a ternary problem-solving machine.
- Svensson, I.-M. et al. (2017). “Period-tripling subharmonic oscillations in a driven superconducting resonator.” arXiv:1707.04093. Three stable phases, measured, in microwave hardware. Why read: proof that three-phase oscillation exists in practice.
- Jang, J. K. et al. (2018). “Synchronization of coupled optical microresonators.” Nature Photonics 12. Two silicon nitride rings locked through 20 m of fibre. Why read: the closest existing physical experiment to the paper’s ring model.
- Kuramoto, Y. (1984). Chemical Oscillations, Waves, and Turbulence. Springer. The standard continuous model of coupled oscillators. Why read: the physical counterpart of the paper’s continuous-phase theorem.
The device layer: multiport optics
- Spagnolo, N. et al. (2013). “Three-photon bosonic coalescence in an integrated tritter.” Nature Communications 4. An integrated tritter, fabricated and characterised. Why read: the central device of the paper already exists.
- Żukowski, M., Zeilinger, A. and Horne, M. A. (1997). “Realizable higher-dimensional two-particle entanglements via multiport beam splitters.” Physical Review A 55. Symmetric multiports as discrete Fourier transforms. Why read: the device model the whole coupler analysis rests on.
- Levy, J. S. et al. (2011). “Harmonic generation in silicon nitride ring resonators.” Optics Express 19. Third-harmonic generation in this exact material. Why read: the basis of device test 4.
The mathematics
- Martin, O., Odlyzko, A. M. and Wolfram, S. (1984). “Algebraic properties of cellular automata.” Communications in Mathematical Physics 93. The ring-algebra method behind the lock law. Why read: the key tool; its Theorem 3.4 is the binary twin of the paper’s ternary lock law.
- Lidl, R. and Niederreiter, H. (1997). Finite Fields. Cambridge University Press. The algebraic foundations used in the proofs. Why read: for anyone who wants to verify rather than trust.
- Berndt, B. C., Evans, R. J. and Williams, K. S. (1998). Gauss and Jacobi Sums. Wiley. The sums behind the even-spreading lemma. Why read: the mathematical engine of the strongest ternary result.
- Gluck (1990); Hiramine (1989); Rónyai and Szönyi (1989). Classification of planar functions on prime fields. Why read: the cited step for the count of uniform windows; independently verified by enumeration for small primes.
- Kumar, P. V., Scholtz, R. A. and Welch, L. R. (1985). “Generalized bent functions and their properties.” JCTA 40. The known class the uniform windows belong to. Why read: connects the paper’s results to established coding-theory literature.
- Wootters, W. K. and Fields, B. D. (1989). “Optimal state-determination by mutually unbiased measurements.” Annals of Physics 191. Quadratic phases at prime dimension, from the measurement side. Why read: a second, independent field that arrives at the same structure.
The ternary frame
- Konstapel, J. (2026). The Vacuum.Net Theory: Foundational Paper, 8th ed. Constable Research, Leiden. The theoretical frame: crossings, trits, windows. Why read: the motivation for choosing three, kept deliberately separate from the proofs.
- Konstapel, J. (2026). Beyond Energy, ed. 2. Constable Research, Leiden. Contains the earlier ring-lock claim. Why read: to see which earlier statements are confirmed, and which are generalised.
- Brusentsov, N. P. et al. “Development of ternary computers at Moscow State University.” Computer Museum, Moscow. Setun: a balanced-ternary computer, 1958–1965. Why read: ternary computing is not new; light makes it fast.
- International Energy Agency (2025). Energy and AI. Paris. Data-centre demand of ~415 TWh in 2024, ~945 TWh expected by 2030. Why read: the energy pressure that makes optical, passive components commercially attractive.