Page 16 - IJAMD-1-2

International Journal of AI for
Materials and Design
Sustainable electronics using AI/ML

the absence of standardized characterization parameters are primarily gathered manually from experiments or
further complicates the integration of existing data into literature, but ML, particularly through widely used
regression models. Variability in how data are collected and large language models, could aid in streamlining the
reported reduces the reliability and comparability of the data collection process. In addition, leveraging existing
data. Due to these challenges, the current focus of studies datasets for enhanced algorithmic refinement, including
has been on classification techniques where sufficient data the application of data augmentation methods to increase
are available. We believe that enhancing data collection dataset size, could contribute to greater efficiency.
methods and standardizing parameter reporting will be Collaborative sharing of domain expertise also offers a
crucial for future research. This will eventually enable the means to alleviate data scarcity, avoiding duplication in
effective use of regression techniques in the ML-aided data collection efforts and thereby minimizing costs and
design of materials. resource utilization. 70

4.2. Challenges in ML for polymer degradation 5. AI-driven material design
Recent QSAR studies have primarily aimed for accuracy, In the 21 century, the primary focus for material
st
often at the cost of transparency, employing models such scientists is to strike a balance between performance
as SVM, GNN, and kNN, which are mentioned above. and sustainability in materials. In general, the design
Although it is established that integrating data and using of materials is influenced by their function and life
ensemble analysis can enhance the reliability of QSAR cycles within specific environments. For example, when
models, they often encounter issues with uncertainty incorporating electronics into biological contexts, factors
for several reasons. For instance, QSAR models such as material biocompatibility, device implantation
may produce false correlations due to errors in the method, and overall design integration must be carefully
experimental process or may not fully capture the data considered. Molecular design serves as a method to
91
characteristics because of the limited size of training achieve the desired functionality.
datasets. In addition, these models inherently require In the realm of polymer chemistry, key factors
the generation of suitable features for training, making that influence their functionality include composition
feature selection a complex task. For example, certain (such as monomer sequence, chain length, dispersity,
structural features of molecules, such as halogens, chain end functionality, side chain type, and backbone)
branching, and nitro groups, have been shown to increase and topology (such as branching, regioregularity, and
biodegradation time, whereas others, such as esters, tacticity). The relationship between polymer composition
amides, and hydroxyl groups, have the opposite effect. and topology spans various chemical scales, from the
However, these structural features cannot be universally molecular to the macromolecular level. Therefore, the
92
applied to represent both readily biodegradable and not molecular structure and polymer composition determine
readily biodegradable molecules. 89
their mechanical, optical, and electrical characteristics.
Designing for multiple properties, particularly when Even slight adjustments in the composition of polymers
considering degradation behavior, is becoming more can directly affect the characteristics of macromolecules,
imperative yet challenging, as optimizing one property especially their flexibility and water solubility. For
often entails compromising others. Other significant example, after examining a collection of 51 low-band-
hurdles in polymer informatics involve developing gap polymers, Roth et al. identified some major trends
93
representations that account for stochasticity, acquiring in the stiffness (tensile modulus) and ductility (crack-
larger datasets, and further exploring retrosynthetic onset strain); wherein the presence of fused rings along
design methodologies. Moreover, the absence of the backbone tends to reduce ductility and increase the
90
standardized characterization for parameters such as modulus. However, the opposite impact is experienced
degradation time and biocompatibility poses a significant by branched side chains in the material. The side chains
challenge that is difficult to overcome in ML-assisted can act as solubilizing agents by introducing functional
designs. Even though only one or several interesting groups or interfering with crystallization. In addition,
86
93
parameters of the original system are often focused, a there have been reports adjusting the polymer’s physical
more systematic treatment is required to understand the characteristics, such as absorption, emission, energy level,
hierarchical relationships by appropriately integrating molecular packing, and charge transfer by engineering the
experimental chemistry, simulations, and data science. side chain. For instance, conjugated polymers are often
In the future, the establishment of additional pertinent insoluble in organic solvents, making them difficult to
databases is necessary to enable easy access to extensive process into thin films. To enhance solubility in nonpolar
datasets. Furthermore, the data currently employed liquids, branched alkyl chains, instead of linear ones, are

Volume 1 Issue 2 (2024) 10 doi: 10.36922/ijamd.3173