Low-e coatings: First designs with GenetiCode
We can now produce layers of silver and 3 oxides. Although we plan to have a larger variety of materials, we are keen to try if we could produce reasonable thin film products with these materials already.
Soon we will get some large area coating equipment with targets like the ones used in our small DD1 device. We could then produce low-e and solar control coatings for architectural glass, for example. Although we know already how these kind of coatings are designed in principle, we will use a new approach of automatic coating design. Our favourite software developer M.Theiss Hard- and Software has just released a new design tool based on a genetic algorithm. The software is called GenetiCode and promises completely automatic thin film design. Since we do not have too much experience with coating design we will just try how it works and what we can get.
From M.Theiss Hard- and Software we received a starting configuration which computes a reflectance spectrum in the infrared (1500 ... 2500 nm wavelength) and a transmittance spectrum in the visible (400 ... 700 nm). In both spectral ranges the target spectrum is set to 1: CODE (which is working inside GenetiCode) will have a low fit deviation if both the IR reflectance of the coating and the transmittance in the visible are high. The coating is placed on top of a 1 mm microscope slide.
GenetiCode optimizes materials and thicknesses. You have to tell the program how many deposition steps you have in your equipment, and what materials are available in each deposition step. Every deposition step is assigned to a layer in a CODE layer stack. We start our design attempts with a single layer, and then we try to improve the design by adding layer by layer. For each layer we allow all our 4 materials to occur, i.e. it can be a silver layer or a layer of oxide A, B or C.
1-layer design
Using just one layer, we do not expect good results, but we start with this simplest choice to practise a little bit and to appreciate the more powerful multilayer solutions that are to follow. The possible materials in the only deposition step are the following:
A-oxide (thickness range: 3 ... 100 nm)
B-oxide (thickness range: 3 ... 100 nm)
C-oxide (thickness range: 3 ... 50 nm)
Ag (thickness range: 3 ... 20 nm)
The genetic algorithm of GenetiCode requires that you let it run several times. In every run, a certain number of designs (i.e. choices of materials and thicknesses) is randomly generated. Then this population develops from generation to generation creating child designs with inherited properties (in most cases), but also new features created by mutation. Good designs have a higher chance to inherite to the next generation, and the quality of the best design increases in many cases from generation to generation. The evolution is followed for a pre-defined number of generations, and finally the best design is taken as the result of the method.
In this 1 layer case the result in all runs is the same, and it is achieved in a very short time. The program suggests to use a silver layer of 7.4 nm thickness:
The quality of the design is expressed in the quantity fitness which is 1/(0.0001 + deviation) where deviation is the CODE fit deviation. If the fit deviation is 0 the fitness is 10000, if the deviation is large the fitness is close to 0.
This poor single layer design has a fitness of 7.2 which will be improved by adding more layers (i.e. do more deposition steps).
2-layer design
Let's invest in a second deposition step to improve the coating. Also here only a few generations are needed to get the final result which the program cannot improve any more:
As we hoped, the fitness is increased to 16.1: Clearly both the IR reflectance and the transmittance in the visible are higher compared to the first 'design' above.
3-layer design
We continue and add another deposition step. This time the achieved fitness is 24.4, and GenetiCode has invented the 'standard' low-e layer stack oxide/metal/oxide:
Due to the oxide layers enclosing the silver a high transmittance in the visible is achieved. Compared to the 2-layer case the Ag thickness is significantly higher which results in a higher IR reflectance. Note that C-oxide (our oxide with the highest refractive index) was selected to be the oxide between substrate and silver layer.
4-layer design
Adding another layer gives GenetiCode the freedom to develop a new concept: The high IR reflectance is made by two Ag layers with an oxide spacer layer in between. The fitness is significantly increased to 52.3:
The total Ag thickness is now more than 14 nm, and the transmittance is above 0.8 in the whole visible spectral range. This kind of double silver layer is used in several commercial coating products at present, and we are happy and satisfied that the GenetiCode software can find such a solution so easily.
5-layer design
The silver double layer coating is re-fined a little more using a fifth layer. The largest fitness that we find with 5 layers is 60.1:
7-layer design
Finally we try a 7-layer stack - just to see if GenetiCode gets another good idea. The best solution with a fitness of 67.0 is obtained by further splitting of the silver layer, now into three layers:
Note the very high IR reflectance which is due to the 19.9 nm silver in the coating!
After these design exercises we are convinced that GenetiCode will be a useful tool for our work. The future will show if we can really produce some of the developed coatings. As far as we know we will have to protect the silver layers in the stack by additional thin films which will change the design conditions a little.