Page 13 - IJAMD-1-2

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

such as PPC and PVA have demonstrated high dielectric 4. ML modeling
57
constant and utilized in the organic field effect transistors
and circuits; these polymers showed instant degradation The main steps in a traditional ML modeling process
via enzymatic degradation in the former and through include data collection, characterization, model training,
70
hydrolysis process in the latter. 58-60 Both the synthetic and model evaluation. Data collection is a crucial stage
in the ML workflow, essential for ensuring accurate
polymers exhibit excellent degradability and as a result, predictions. High-quality and quantity datasets are vital,
their composites have been developed for a variety of as they directly affect the potential performance of the
applications, which can be found elsewhere. 61,62 Besides, ML models. The second step, also known as feature
synthetic and elastomeric polymers, natural polymers extraction, involves generating and selecting descriptors
66
such as cellulose, silk fibroin, keratin, jute, and from the original data. This transformation is essential as
63
64
65
bamboo exhibit high k values and are used in organic it impacts the quality and interpretability of the resulting
67
thin film transistors. In addition to organic dielectric model. It is important to recognize that the attributes
materials, new and innovative electrolytes are constantly of the data determine the limit of maximum likelihood,
being tested for application in biodegradable organic with the algorithm only capable of approximating this
transistors. Incorporating biodegradable electrolyte upper limit. Algorithm design and model training, the
71
materials into organic transistors introduces a novel range third step, is pivotal in ML, where different algorithms
of biological capabilities, enhancing their applicability may yield varying results on the same dataset. ML
in upcoming wearable, implantable, and electronic skin primarily comprises two types of learning: supervised
applications. 68,69 and unsupervised. Supervised learning involves fitting
3.5. Summary of functional biodegradable materials a model to labeled data to predict outcomes, while
unsupervised learning detects patterns in unlabeled data,
This review focuses on recent research concerning transient employing techniques such as clustering and dimension
materials, covering metals, polymers, and semiconductor reduction. The algorithm design and model training
72
materials. In the realm of metals, researchers are exploring step can be challenging because different algorithms
the potential of Mg, Zn, Fe, W, and Mo as soluble metals, may perform variably on the same dataset, choosing
each offering specific degradation characteristics suitable and optimizing the right hyperparameters is crucial for
for various applications. In fact, Mg and Zn degrade achieving optimal prediction performance. The last
73
quickly, whereas W and Mo degrade at slower but more step involves utilizing various metrics to gauge their
predictable rates. In the domain of polymers, both performance in regression and classification models.
natural and synthetic varieties are investigated for their The efficacy of ML model is based on their capability to
biodegradability. Natural polymers such as cellulose accurately forecast unknown data and effectively fit the
and silk, polysaccharides such as starch and gelatin available data points. 74
demonstrate excellent potential for transient electronics
due to their biocompatibility and degrading capabilities, 4.1. ML techniques in biodegradability
wherein synthetic polymers such as PGA, PLA, PLGA, Determining the biodegradability of chemicals without
PCL, and PVA offer tunable degradation rates, allowing relying on costly tests is both ecologically and economically
for tailored transient behavior. PLGA, a blend of PLA advantageous. Quantitative structure–activity relationship
and PGA, offers controlled degradation, while PCL and (QSAR) models offer potential in this area. The QSAR
PVA are notable for their slower and faster degradation prediction system is designed for classifying biodegradation
rates, respectively. In the field of semiconductors, the datasets without the need for actual chemical experiments.
quality and characteristics of semiconducting materials QSARs are mathematical models that predict the physical,
are vital for electronic devices. Research in degradable chemical, and biological properties of substances based
electronics focuses on materials such as Si-based, metal on their molecular structures. These systems have gained
oxides, and OSs. Both inorganic and organic dielectric attention as numerous countries have updated their
materials are important components of electrical devices environmental policies to reduce the use of environmentally
in the area of dielectrics. Inorganic dielectrics such harmful, non-biodegradable substances. For instance,
as MgO, SiO , and Si N are used in FETs and other European legislators incorporated chemical persistence in
2
3
4
applications, demonstrating complete degradation in the registration, evaluation, and authorization of chemicals
deionized water. Organic polymers such as PLA, PVA, and for chemical evaluation. These regulations utilize QSAR
PMMA are popular due to their commercial availability models for assessing chemical risks. The most recent,
and biodegradability, with PPC and PVA showing high effective, and commonly used ML techniques in predicting
dielectric constants and facile processing. biodegradability include classification and regression

Volume 1 Issue 2 (2024) 7 doi: 10.36922/ijamd.3173