---
title: "MIT builds the photonic chip on silicon, then removes almost all the silicon"
locale: "en"
url: "https://irz.fr/en/articles/photonic-chip-remove-silicon-flexible-en"
markdown_url: "https://irz.fr/en/articles/photonic-chip-remove-silicon-flexible-en.md"
category: "tech"
tags: ["photonics", "semiconductors", "fabrication", "flexible", "MIT"]
published_at: "2026-09-23T09:00:00.000Z"
author: "Arthur Lacoste"
translation: "https://irz.fr/fr/articles/puce-photonique-silicium-retire-flexible-fr.md"
---

# MIT builds the photonic chip on silicon, then removes almost all the silicon

The process keeps the wafer rigid through the steps that need precision and foundry tooling, then removes the original support and transfers the optical layers onto transparent film. Flexibility arrives at the end.

A flexible photonic chip begins with an awkward manufacturing problem. The machines best at producing tiny, repeatable optical structures want a **rigid, flat, carefully controlled wafer**, while the applications Jelena Notaros' group is targeting want almost the opposite: something thin, transparent and willing to follow a curve.[1](https://doi.org/10.1364/OPTICA.589726)[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)

The new MIT and NY Creates platform does not try to make the entire semiconductor line flexible; it keeps rigidity for the steps that benefit from it, then removes that rigidity at the end.

Fabrication starts on a **300 mm silicon wafer**, using foundry-compatible semiconductor tooling. Waveguides and optical layers are deposited and patterned on that rigid base. Only afterward does the team bond a temporary silicon carrier, flip the stack, remove the original silicon substrate until only a few microns of functional material remain, attach transparent polyester film and finally debond the temporary carrier.[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)

Flexibility is therefore not a fabrication condition: the device is **fabricated like a foundry chip and then released from its manufacturing support**.

## Wrong support

Integrated photonics has benefited enormously from borrowing the machinery of microelectronics.

On a 300 mm wafer, a foundry can repeat nanoscale structures with a level of uniformity that an improvised process on a soft membrane would struggle to match. The 2026 paper frames its main contribution accordingly: a transparent and mechanically flexible wafer-scale silicon-photonics platform compatible with large-scale fabrication.[1](https://doi.org/10.1364/OPTICA.589726)

The traditional support becomes the awkward part. Bulk silicon is almost irrelevant underneath a rack-mounted optical transceiver, yet its opacity and rigidity dominate the problem when the same optical layer is meant to follow skin, a windshield or a visor.

The Notaros group had already attacked half of this contradiction in 2024. A *Scientific Reports* paper described its first mechanically flexible photonics platform fabricated across a 300 mm wafer, with visible-light waveguides, passive devices and repeated mechanical testing.[3](https://www.nature.com/articles/s41598-024-61055-w)

That work solved one scaling problem: **make flexible photonics using a real wafer-scale process**, rather than producing individual devices one at a time. The *Optica* paper published in 2026 adds the second property that changes the application space: **transparency**.[1](https://doi.org/10.1364/OPTICA.589726)[4](https://web.mit.edu/notaros/www/publications.html)

## Keep the wafer

The process intentionally begins in a conventional way, with the researchers depositing and patterning tiny optical waveguides on a rigid silicon substrate using standard semiconductor manufacturing equipment at NY Creates' Albany NanoTech Complex.[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)

The apparent contradiction disappears once the manufacturing substrate and final substrate are treated as two separate jobs.

A rigid wafer gives the equipment exactly what it expects: a known, flat surface that automated tools can handle, align, coat and pattern. Instead of redesigning those machines around a floppy sheet, the team **postpones flexibility**.

> **Two supports, two jobs**
> - flatness, robotic handling, deposition and lithography during fabrication: Silicon wafer
> - holds the membrane while the original substrate is thinned and flipped: Temporary carrier
> - becomes the permanent thin, transparent and flexible support: Polyester film
> - the functional material the process tries to preserve across support changes: Optical layers
> The platform separates the material useful to manufacturing from the material useful to the final device.

## Flip the stack

Once the optical layers are fabricated, the researchers bond a **second, temporary silicon wafer** on top, flip the assembly and expose the original substrate so it can be removed.[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)

This is much more delicate than simply peeling off a chip. Almost all the original silicon thickness has to disappear without cutting through or damaging the ultrathin optical layers the process is trying to preserve.[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)

The team combines removal methods. Industrial thinning takes away most of the material, then a more selective chemical etch handles the final portion so the remaining layers survive.[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)

What remains is a membrane only **a few microns thick**, containing the oxide and waveguiding layers. MIT describes it as less than one tenth the thickness of a human hair.[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)

Thinning one tiny sample would already be delicate; here the removal has to stay controlled across a **300 mm wafer** without producing cracks, ripples or uncontrolled thickness variation.

## Wafers bow

A wafer is never perfectly free of stress. Deposited films, thermal cycles and material mismatches can leave internal stress that makes it bow. While the substrate remains thick, the plate resists; as silicon disappears, that stiffness disappears too and accumulated strain becomes much harder to ignore.[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)

MIT describes a very concrete failure mode: if strain is not managed during the flips, the wafer can develop surface ripples or **shatter in the fabrication line**.[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)

The researchers therefore kept the relevant process temperatures at **500 °C or below** to control thermal stress.[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)

That detail may matter more than the spectacular photograph of a transparent sheet being bent by hand. Final flexibility depends on a sequence that, for most of its duration, has to behave like a disciplined rigid-wafer process.

## Bond transparency

With the original silicon gone, the membrane still needs a usable permanent support. The team attaches a **thin, transparent polyester film** to the remaining layers with adhesive.[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)

The film becomes the permanent carrier, making it possible to debond the temporary silicon wafer from the opposite side.[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)

The result is a very thin sheet whose optical layers retain their geometry, while its mechanics are now governed by a membrane and polymer film rather than a thick silicon slab.

The division of labour is unusually clear: silicon is the support used to **manufacture**, while polyester is the support used to **live in the application**.

There is no physical rule saying those two jobs need the same material. Traditional processing merely encourages that assumption because the starting substrate naturally survives to the end.

## The 2024 step

The 2024 paper makes the development path unusually visible, because that first generation already targeted wafer scale before transparency was added.

That first 300 mm platform was already mechanically flexible and compatible with wafer-scale CMOS-style fabrication.[3](https://www.nature.com/articles/s41598-024-61055-w) It operated at visible wavelengths using silicon-nitride waveguides and demonstrated coupling, routing and passive devices.

Its optical losses were not records: the paper measured **12.1 dB/cm** for 300 nm waveguides and **9.4 dB/cm** for 400 nm waveguides at 632.8 nm.[3](https://www.nature.com/articles/s41598-024-61055-w)

The authors explicitly noted that rigid visible-light platforms could report lower values and listed routes to improvement including annealing and a move toward LPCVD silicon nitride.[3](https://www.nature.com/articles/s41598-024-61055-w)

That honesty is useful when reading the 2026 work. The platform is not a simultaneous victory on every metric; it buys **a new physical form factor** while trying to retain enough optical quality and foundry compatibility to matter.

## Two thousand bends

The same 2024 work subjected one flexible chip to a particularly readable protocol: 500 bends around a 2-inch cylinder, then 500 at 1.5 inches, 500 at 1 inch and 500 at 0.5 inch.[3](https://www.nature.com/articles/s41598-024-61055-w)

Across the full sequence, the published experiment accumulated **2,000 bends with no noticeable optical degradation**.[3](https://www.nature.com/articles/s41598-024-61055-w)

At **0.25 inch**, the chip eventually failed when a crack crossed the waveguide and transmitted optical power dropped sharply.[3](https://www.nature.com/articles/s41598-024-61055-w)

For the 2026 transparent platform, MIT again reports thousands of bends around cylinders of different diameters with no performance loss down to roughly the diameter of a small screw; repeated bending around something about the size of a toothpick eventually starts degrading the device.[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)[5](https://www.photonics.com/Articles/Wafer-Scale-Fabrication-Process-Brings/p15/a72587)

The two experiments should not be blended as though they used the same sample or exactly the same material stack. They show continuity in method: **measure where flexibility stops being free**.

> **Two generations of flexible wafer**
> - first mechanically flexible 300 mm platform, characterized at visible wavelengths: 2024
> - down to 0.5 inch with no noticeable optical degradation in the published protocol: 2,000 bends
> - the platform also becomes transparent while retaining 300 mm scale: 2026
> - add more complex components and improve efficiency and transparency: Next
> Same research line, separate papers: 2024 metrics should not be silently assigned to the 2026 stack.

## Look through it

Transparency creates a problem that a simple wearable patch can ignore.

A chip in front of an eye might transmit plenty of light and still be unpleasant if it scatters, blurs or distorts the scene. The team therefore used what MIT calls a **bionic eye** to look through the chip and evaluate how the viewed image changed.[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)

The reported result remains qualitative but addresses the right failure mode: low haze and no noticeable image distortion in that test.[1](https://doi.org/10.1364/OPTICA.589726)[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)[5](https://www.photonics.com/Articles/Wafer-Scale-Fabrication-Process-Brings/p15/a72587)

The *Optica* paper describes the contribution in terms of **low haze and low distortion**, rather than claiming perfect invisibility.[1](https://doi.org/10.1364/OPTICA.589726)

That distinction matters. A layer intended for a visor or transparent display has to be evaluated as an optical element through which an image is seen, not merely as a chip that still guides light internally.

## What is missing

The 2026 platform is not a transparent pilot helmet ready for production.

MIT explicitly says the team now wants to add more complex components and functionality, improve waveguide efficiency and further improve transparency performance.[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)

The applications it names, discreet body-worn sensors, curved AR displays and pilot visors, are **directions enabled by the platform**, not products demonstrated in this paper.[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)[5](https://www.photonics.com/Articles/Wafer-Scale-Fabrication-Process-Brings/p15/a72587)

That is exactly what a fabrication platform is supposed to do: change what later teams can build even before the final system exists.

## Two timelines

The MIT process is interesting because it separates two moments that manufacturing often bundles together.

During **fabrication**, the chip benefits from rigidity, flatness, chemical and thermal compatibility, and a format that tools can reliably grab and align.

During **use**, it may instead benefit from transparency, low mass, curvature and contact with a moving surface.

Choosing one material and asking it to win both competitions is not obviously rational. The Notaros group instead moves the functional layers between supports.

This logic reaches beyond photonics. It appears whenever the material best for manufacturing is poor for use: electronics on skin, conformal sensors, circuits in textiles and components wrapped around non-planar geometry.

One option is to make the whole manufacturing line compatible with the final support. Another is to fabricate **somewhere else**, then transfer what matters.

## The real flexibility

The photograph of a transparent sheet bent between two hands points attention toward mechanics, while the manufacturing result sits mostly in compatibility with wafer-scale tooling.

A flexible membrane made manually over a few square centimetres may remain a laboratory curiosity. A platform that retains a **300 mm wafer**, foundry-grade tools and a repeatable process changes the scale of the problem.[1](https://doi.org/10.1364/OPTICA.589726)[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)

That is also why the NY Creates collaboration matters as much as the material stack. Wafer processing happened on semiconductor-manufacturing infrastructure, not only on a bespoke setup built around one sample.[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)

The team hopes to make the platform available to other researchers through that infrastructure.[2](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)

Nothing yet guarantees that a future eyewear or medical-device manufacturer will use this exact stack; what has already changed is the question the manufacturing process allows them to ask.

It is no longer only: **can a photonic chip bend?**

It can now be: **can it be fabricated using the machines that already know how to make complex wafers, then choose its body at the end?**

The MIT process answers yes to that second question. That answer may matter more than the bend itself.

## References

1. [Tal Sneh et al., Transparent and mechanically flexible wafer-scale silicon-photonics fabrication platform, Optica 13(9), 1751-1758, 2026](https://doi.org/10.1364/OPTICA.589726)
2. [MIT News, Fabrication platform could enable flexible, transparent next-generation photonic chips, September 3 2026](https://news.mit.edu/2026/fabrication-platform-could-enable-flexible-transparent-next-generation-photonic-chips-0903)
3. [Milica Notaros et al., Mechanically-flexible wafer-scale integrated-photonics fabrication platform, Scientific Reports 14, 10623, 2024](https://www.nature.com/articles/s41598-024-61055-w)
4. [MIT Photonics and Electronics Research Group, Publications](https://web.mit.edu/notaros/www/publications.html)
5. [Photonics Spectra, Wafer-Scale Fabrication Process Brings Next-Generation Displays, Wearables Into Focus, September 2026](https://www.photonics.com/Articles/Wafer-Scale-Fabrication-Process-Brings/p15/a72587)