June 16, 2020

20 Years of Materials Modeling

A quick recap of the past and future outlook

A quick recap of the past and future outlook

The Two Big Events of 2020

Right before we went on lockdown because of COVID-19, Schrödinger, provider of software and services for molecular modeling in pharmaceutical and materials, went public with a market cap of $1.5B (SDGR) with a current market cap nearing $4B [1]. The company was started in 1990 and the road to the IPO took 30 years.

SDGR listing on Nasdaq in February 29020

Materials Studio, another materials modeling suite and a competitor of Schrodinger, celebrated its 20th anniversary [2]. The core of the software engine is likely closer in age to Schrodinger as its 2000 founding was following a merger of multiple companies into Accelrys, which later became Biovia.

Materials Studio Interface

Although Advanced Materials is a trillion-dollar industry [3], these leading materials modeling companies took decades to develop and scale.

What will the next 20 years look like for the industry?

The 1.0 Approach is Too Slow

Schrodinger and Materials Studio both represent the first wave of modeling software for materials that emerged in the 1990s. Although a strong success, the 30-year path to the IPO is not particularly exciting for venture capitalists today seeking 8–10-year time-frames for exits and growth rate rivaling that of Facebook and Slack.

What can we do to make the industry go faster? Given the technology advances, collaborative, open, and data-centric alternatives can accelerate the timing. Materials Science is an extremely complex discipline and no single company can address all of the challenges alone, especially behind closed doors. Preserving intellectual property is feasible with properly designed secure digital collaboration platforms. Additionally, these types of environments will enable faster decision-making and will speed up the work.

Open alternatives from academia are plentiful but are often duplicated, tied to the agenda of a specific funding source (i.e., governmental program) instead of industrial needs. That’s why such initiatives often fail to address the needs of and face significant skepticism from the industrial players.

In-house (software) developments within Global 2000 are by-and-large limited to today’s necessities only and are an absolute nightmare to maintain in the long term. With the ever-changing technology landscape, writing good software is already challenging. Most in-house software development initiatives in materials modeling will likely fail or continue to be extremely inefficient and expensive even for Global 2000 companies with deep pockets whose core business is not in software.

Lastly, the emergence of Data-driven Science and Artificial Intelligence (AI) calls for a completely different view, embracing the long-term value of data over the tools and workflows only. The latter have to evolve to allow for intuitive data collection and organization for teams with multidisciplinary backgrounds in materials/chemistry and data science and machine learning (ML).

Data-driven Science as the Fourth Paradigm emerging in the 21st century

The 2.0 Approach is Emerging

Well-organized Standards are critical for any kind of efficient work, both internally within an organization and when facilitating a collaboration, whether internal or external. Internally the standards promote the data-centric approach and create a repository of knowledge accessible to future generations of scientists and to AI/ML

A proverb says “Two heads are better than one”. Collaboration is what makes the enterprises execute their core functions, and R&D is often one of them. Software engineers and Data Scientists demonstrate this very well with collaborative platforms like GitHub, Azure ML Studio, for example. For computational materials science, the analogy is very clear, however, the real-world implementation is much more complicated.

Because of its function and complexity, materials R&D is perceived by the corporates as (1) producing sensitive information that must have restricted access, (2) being too complex to organize in a software domain relying instead on the scientists to “figure out some sort of structure”. Both of these are valid concerns, however, as COVID-19 demonstrated, they are only blocking the progress in the world where digital is optional, and this optional world is now gone. Facilitating open-access without endangering intellectual property and embracing community-curated standards is the future.

At the beginning of the materials modeling 1.0 era in the 1990s, the average computational system was at least 2¹⁵ (~ 32,000 times) less powerful than what we have now. With the availability of specialized computing like massively-parallel CPUs, GPUs, TPUs, and soon QPU (quantum computing), the capabilities of what we can do in simulations will continue to expand. There is no doubt that in the next decade will prove to be an inflection point in materials modeling — for biopharmaceutical drug development, new energy sources, sustainable materials, agile manufacturing, and many other critical sectors of our everyday lives.

The Time is Now

This year Netflix is spending 3–4x more money on generating new content [4] than Taiwan Semiconductor Manufacturing Company (TSMC) is on their R&D [5]. Netflix has millions of people glued to their digital screens and TSMC produces most of the world’s computer chips that are on these devices. Given the demand for improved computing, it is quite conceivable that these two numbers are switched with R&D spends outpacing digital content spend. Expect this for not only the semiconductor industry, but for pharmaceuticals, plastics waste recycling, energy, aerospace, and many other sectors.

To address the challenges of today (COVID-19, climate change, pollution, etc.) materials science will be foundational. Yet, we still predominantly approach modeling like it’s 1999. Solutions will need to manage far larger scales of data than we compute today and collaboration will be a path for accelerated development in educational and industrial areas. 8,000,000,000 N95 masks are not recyclable and these plastic-based articles are already showing up as COVID-waste in our oceans.

We’re not facing our last pandemic. Every aspect of our life needs a material change. And we don’t have 30 years to create the next unicorn. As 2020 has clearly shown, global crises are accelerating and affecting everyone. Advancing digital technologies, and materials modeling, in particular, is fundamental in making our lives safer, better, and greener.

— —

Team Exabyte.io: Marta Bulaich, Timur Bazhirov.