Skip to content
Matmerize

Designing Polymers for Fuel Cells and Batteries

Developing new polymers is essential to meet the performance demands of next-generation fuel cells and
batteries. In these applications, the primary requirement is achieving sufficiently high proton conductivity for
polymer electrolyte membranes (PEMs) in fuel cells, and high lithium-ion conductivity for solid polymer
electrolytes (SPEs) in lithium-ion batteries. Beyond ionic conductivity, these polymer layers must also exhibit
electrochemical stability, mechanical strength, thermal stability, and other critical properties. Moreover, to
surpass current benchmark materials—such as Nafion in fuel cells—new polymer candidates must also be
synthetically accessible to ensure practical viability.
PolymRize™ is the only standardized software platform dedicated to polymer informatics, integrating proprietary
polymer fingerprinting schemas, advanced machine learning algorithms, generative design tools, and domain-
specific insights. It enables the efficient and targeted design of polymers for a wide range of applications. In this
work, we demonstrate its utility in the context of emerging energy storage technologies.

Approach

Designing polymers to meet target specifications
using informatics approaches follows a structured
workflow (illustrated in Figure 1 for the example of
PEM design for fuel cells):
• Step 1: Define the "victory condition" by
specifying target properties and the desired value
ranges.
• Step 2: Ensure the availability of rapid, predictive
models for these properties. If such models are
not readily available, relevant data must be
gathered—either from experiments or
simulations—and machine learning models
trained accordingly.
• Step 3: Construct the chemical search space, i.e.,
a curated or generated set of candidate polymers.
• Step 4: Apply screening criteria to the search
space using the predictive models to identify
promising candidates that meet the target
property thresholds.
For applications such as PEMs (for fuel cells) and
SPEs (for batteries), PolymRize™ includes pre-built
property prediction models for virtually all relevant performance metrics (Figure 2). As a result, once the
screening criteria—i.e., the target property values—are defined (Step 1), Step 2 is already addressed through these
pre-trained models. This allows users to seamlessly proceed to Steps 3 & 4.
Figure 1. Workflow for Designing New-to-the-World Polymers:
From Property-Based Screening Criteria to Synthesizable
Candidates
We demonstrate this 4-step process for the example problem of
designing a new PEM membrane [1] that may surpass Nafion, the
standard material used today.
The most critical property of PEM materials is proton conductivity.
This property is inherently linked to the material’s ability to absorb
water, since protons are transported through hydrated channels
within the polymer matrix. Thus, water uptake is strongly
correlated with proton conductivity. Leveraging a curated database
containing measurements of both properties, we have developed
best-in-class predictive models using our proprietary multi-task
(multi-target) co-learning algorithm, which captures and utilizes
the underlying correlations between these properties. Figure 3
illustrates the predictive performance of these models.
PolymRize™ includes pre-built models for all
other essential PEM properties—including O₂
and H₂ gas permeability, electronic band gap,
glass transition temperature (Tg), thermal
decomposition temperature (Td), and
Young’s modulus—making it a
comprehensive tool for polymer design in
electrochemical energy applications.
Suitable target value criteria were established
for each of these properties [1], and a
candidate screening set was assembled,
comprising over 30,000 previously
synthesized homopolymers and copolymers—
none of which had been previously considered as PEM materials. Applying the PolymRize™ predictive models and
screening criteria to this candidate set, we identified several polymers that are projected to meet or exceed the
performance of Nafion, the current industry standard.
Figure 4 shows the chemical
structure of one such top-performing
polymer alongside Nafion, as well as
the predicted proton conductivity
and other key property attributes.
The selected polymer clearly
surpasses Nafion in proton
conductivity across a broad humidity
range typical of fuel cell operation.
Since these candidates have already
been synthesized for other
applications, they may now be
repurposed and tested for lab-scale
Figure 2. PolymRize pre-built property stock
models relevant for fuel cell and battery
applications.
Figure 3. ML predictor for proton conductivity and water uptake.
Figure 4. Proton conductivity of newly designed polymer-PEM1 and Nafion (left),
chemical structure and other attractive attributes of PEM1 (right)
PEM performance, accelerating the path to experimental validation.
PolymRize™ enabled rapid and large-scale exploration of the chemical space, virtually screening tens of
thousands of polymers and performing multi-property optimization in a matter of seconds. This ability to sweep
massive swathes of the search space at minimal cost is essential to access and evaluate the full breadth of polymer
candidates.
Such high-throughput, informatics-driven screening streamlines R&D by intelligently identifying the “next best
experiment.” These experiments can then be conducted, the resulting data fed back into the platform, and a new
predictive model trained—further refining recommendations. Through this iterative, closed-loop approach,
PolymRize™ accelerates convergence toward high-performing target materials in a rational, cost-effective, and
scalable manner.
A similar strategy may be adopted for any problem for which property-based screening criteria, rapid and robust
property predictors (or a pathway to create them), and a capability to generate or define the search space of
synthesizable polymers. PolymRize™ supports over 50 property models, multi-property learning, automatic
generation of candidate polymers and multi-property optimization through its proprietary algorithms [2].
[1] Huan Tran, Kuan-Hsuan Shen, Shivank Shukla, Ha-Kyung Kwon, Rampi Ramprasad, Informatics-Driven Selection of Polymers for Fuel-
Cell Applications, J. Phys. Chem. C 127, 977 (2023).
[2] Christopher Kuenneth, Arunkumar Chitteth Rajan, Huan Tran, Lihua Chen, Chiho Kim, and Rampi Ramprasad, Polymer informatics with
multi-task learning, Patterns 2, 100238 (2021).