Technical Issues - Will AI Eliminate the Need for Optical Coating Engineers?

A FilmStar user recently inquired about AR on Silicon with film materials Si3N4 and SiO2.

Can ChatGPT, the artificial intelligence bot much in the news recently, design optical thin film coatings? Is it time to look for a new job?

"Design a visible (450 to 750 nm) antireflection coating for silicon substrate with two thin film materials having index 2.02 and 1.46 and 10 or fewer layers"

 

Response: An antireflection coating can be designed using thin film interference. One common method is to use a multilayer stack of alternating layers of high and low refractive index materials.

For a silicon substrate with a refractive index of about 3.4, a visible antireflection coating can be designed using two thin film materials with refractive indices of 2.02 and 1.46.

One possible design for a 10-layer antireflection coating is:

Layer 1: High refractive index material (n = 2.02) with thickness of 65.4 nm
Layer 2: Low refractive index material (n = 1.46) with thickness of 38.2 nm
Layer 3: High refractive index material (n = 2.02) with thickness of 56.7 nm
Layer 4: Low refractive index material (n = 1.46) with thickness of 33.5 nm

Layer 5: High refractive index material (n = 2.02) with thickness of 49.5 nm

Layer 6: Low refractive index material (n = 1.46) with thickness of 29.3 nm

Layer 7: High refractive index material (n = 2.02) with thickness of 43.5 nm

Layer 8: Low refractive index material (n = 1.46) with thickness of 25.6 nm

Layer 9: High refractive index material (n = 2.02) with thickness of 38.6 nm

Layer 10: Low refractive index material (n = 1.46) with thickness of 22.3 nm

This coating design will have high transmittance and low reflectance in the visible range of 450 to 750 nm wavelength. However, the thickness and material properties are dependent on the specific application and may require further optimization.

Si index is about 3.4 at 2 µm, not in the visible, but we'll use the fake news value anyway. ChatGPT's 10 layer design is #1 in the graph below. DLS optimization gives curve #2. Noting that ChatGPT gave an excellent result at one wavelength, perhaps in the range was interpreted as somewhere in the range?

ChatGPT + DLS: 73.26H 84.36L 51.35H 34.15L 68.15H 38.08L 56.5H 20.38L 17.35H 94.26L 

Clicking Regenerate response gives a completely different answer:

Response: An antireflection coating for a silicon substrate in the visible range (450 to 750 nm) can be designed using a stack of alternating layers of the two thin film materials, with indices of refraction of 2.02 and 1.46. A common design is the "two-layer quarter-wave stack," in which the thickness of each layer is a quarter of the wavelength of the light being reflected. This design can be used with two to ten layers.

For example, a coating with two layers can be designed as follows:

Layer 1: Material with index of refraction 2.02, thickness = (lambda/4) / 2.02
Layer 2: Material with index of refraction 1.46, thickness = (lambda/4) / 1.46
Where lambda is the wavelength of the light being reflected.

This design is known to provide good antireflection properties in the visible range, with low reflectivity and high transmission. However, the exact thickness of the layers may need to be optimized for the specific application and substrate, as well as taking into account other factors such as the properties of the materials and the manufacturing process.

While better (#4 in the above graph) than the 10 layer solution, note the phrase may need to be optimized. DLS optimization gives #5.

As further Chat GPT Regenerated responses were not useful, we tried "educated guess and optimize". Starting with 5 50 nm HL pairs [(50H 50L)5], we very quickly obtain the following with DLS refinement.

It seems to us that optical coating engineers need not worry (yet) about being replaced by AI. ChatGPT should have responded with a message like "This is beyond ChatGPT's capability. Please consider Ron Willey's Practical Design of Optical Thin Films ZOOM course."

Copyright © 2023 FTG Software Associates
Last updated on February 08, 2023