Skip to content
Matmerize

Design Polymer Dielectrics with Desired Properties

Imagine that you need to design polymer dielectrics for one of many
electronics or electrical applications. Depending on the specific application,
a particular set of properties may be required. Figure 1 shows an example
set of such property-need-based screening criteria for the case of low-loss
capacitor dielectrics that can withstand high temperatures. The challenge
with designing materials that meet multiple property criteria is that, often,
the properties are unfavorably entangled (Figure 2). For instance, while
high values may be desired for two properties, these properties may be
inversely related.
Further imagine that the designed materials must be easy to synthesize.
Matmerize’ PolymRizeTM platform, uses artificial intelligence-based
methodologies and can help design such polymers. Such virtual design
methodologies have been validated recently through collaboration with
synthetic chemists and electrical engineers [1]. This work has resulted in
new previously unknown classes of high-T, low-loss, high-breakdown
strength polymer dielectrics that is also amenable to scale up.

Approach

Designing polymers that meet target criteria using informatics approaches proceeds as follows (Figure 3). First, in
Step 1, one decides on the “victory condition”, namely, determines the target properties and desired values for
those, such as shown in Figure 1. Next, in Step 2, one ensures that rapid predictors of the relevant properties are
available; if not, effort needs to be put in place to procure data for the specific property, and machine learning
models need to be built to predict the property. The sources of data may be from experimental measurements or
Figure 1. Example screening criteria.
Figure 2. How to surpass commercial standard materials?
physics-based simulations. Next, in Step 3, the search space
needs to be defined on which to apply the screening criteria
(in Step 4) to identify suitable candidates that may possess
the desired property values.
PolymRizeTM can perform all the above steps. While Step 1
needs to be determined by the user, all other steps are
automated in PolymRizeTM. For instance, pertaining to Step
2, several pre-built stock models are available (in fact, robust
predictive models are available for all the properties listed in
Figure 1). Even if such a predictive model for a particular
property is not available, PolymRizeTM comes with a “custom
model” building capability which the user may use, in
conjunction with their data, to build their own predictive
models for their exclusive use.
The user will have several options to handle Step 3, namely,
creation of the candidate set (i.e., defining the search space).
The most trivial way to do this is to list up commercially
available (or internally synthesizable) polymers, make
predictions of the relevant properties for this set, and select
those that meet the screening criteria. A second approach is
to use PolymRizeTM’ built-in “Chemical Variants” feature
that starts with one or more user-suggested starting
polymers and produces a set number of variants of those
starting polymers. The variations are selected from our
catalog of synthetically feasible functional group variations
that may be accommodated in the polymer. A third method
to generate polymers is to start with commercially available
monomers and known polymerization templates and
virtually produce a plethora of polymers which, by
definition, will be synthetically accessible.

Results - Accurate and Useful
Predictions

Using the above protocol, several new
classes of high-temperature, high-
breakdown strength polymer dielectrics
for capacitive energy storage have been
designed, via screening of several
thousands of synthesizable polymers.
Figure 4 shows some recent examples of
such new-to-the-world polymers,
validated through experimental synthesis
and testing. These new materials are
compared against presently used
standard materials. Of particular note is
Figure 3. Workflow to design new-to-the-world polymers.
Figure 4. Example designs of high-T, high-energy density polymer dielectrics.
the PONB1 polymer whose performance is shown using a star. This is a record-setting polymer with an
unprecedented energy density at high temperatures. Many of these polymers are currently going through scale up.
PolymRizeTM was thus able to sweep massive swathes of the search space in seconds and handle properties which
tend to be unfavorably entangled. The ability to virtually screen tens of thousands of polymer formulations at
minimal cost, and at scale, is necessary to access and assess the breadth of the chemical-process space. Such a
capability can be used to streamline R&D efforts by intelligently selecting the “next best experiment”, performing
them, adding the new data to the original set, building a fresh model, making recommendations for the next set of
experiments, and thus, iteratively, but efficiently, converging on to the target materials in a rational and
accelerated 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. Multiple properties, multi-property learning, automatic generation of candidate materials
and multi-property optimization can be handled in PolymRizeTM using proprietary algorithms [2].
[1] Chao Wu, Ajinkya Deshmukh, Lihua Chen, Rampi Ramprasad, Gregory Sotzing, and Yang Cao, Rational design of all-organic flexible high-
temperature polymer dielectrics, Matter, 5, 2615 (2022).
[2] Christopher Kuenneth, Arunkumar Chitteth Rajan, Huan Tran, Lihua Chen, Chiho Kim, and Rampi Ramprasad, Polymer informatics with
multi-task learning, Patterns, 2, 4, 100238 (2021).