Molecules That Think: How Indian Scientists Just Changed the Future of AI
By Athira Vijayan | R&D Chemist | know_chemicals | May 2026 | 12 min read
"Your brain has 86 billion neurons that learn through chemistry. Scientists at IISc Bangalore just made a single molecule that does the same thing electrically. The future of artificial intelligence might not be a bigger silicon chip. It might be a better molecule."
Something remarkable happened in a lab in Bangalore in December 2025 — and most people missed it.
Researchers at the Indian Institute of Science published a paper in Advanced Materials — one of the most respected journals in materials science — describing a molecular device that can think, remember, and learn. Not metaphorically. Not as a simulation. Physically. As a function of chemistry.
As an R&D chemist working in Bangalore, this paper stopped me completely. Because what it describes is not just a scientific advancement — it is a signal that the entire architecture of artificial intelligence hardware is about to be rethought. And chemistry is at the centre of it.
🧠 First, Understand the Problem With How AI Works Right Now
To understand why this discovery matters, you need to understand the fundamental tension at the heart of modern artificial intelligence.
Every AI system you interact with today — ChatGPT, image generators, voice assistants — runs on silicon chips. These chips are extraordinarily powerful. They can process billions of calculations per second. But they have one fundamental limitation that has not changed since the first transistor was invented in 1947:
A silicon transistor can only do one thing: switch between on and off. It is either 1 or 0. It processes information, and separately, stores it. It cannot do both simultaneously. And critically — it cannot learn. It can only follow instructions that humans write for it.
Your brain, by contrast, works completely differently. Your neurons do not simply switch between states. They strengthen connections when you repeat something. They weaken connections when something becomes irrelevant. They process and store information in the same place, at the same time, using chemistry — specifically, the flow of ions and neurotransmitters across synaptic junctions.
This is why your brain can recognise your mother's face in a dark room, in under a millisecond, using approximately 20 watts of energy — the equivalent of a dim light bulb. Meanwhile, the AI systems that attempt to do the same thing consume megawatts of electricity across entire server farms.
The fundamental problem is this: we have been trying to build brain-like intelligence on hardware that is nothing like a brain. And for 50 years, the field of neuromorphic computing has been searching for a better solution. In December 2025, a team from IISc found one.
🔬 The Discovery: What IISc Actually Did
The paper, titled "Molecularly Engineered Memristors for Reconfigurable Neuromorphic Functionalities," was published in Advanced Materials on December 9, 2025. It was led by Professor Sreetosh Goswami, Assistant Professor at the Centre for Nano Science and Engineering (CeNSE) at IISc Bangalore, in collaboration with Professor Navakanta Bhat and an interdisciplinary team spanning chemistry, physics, and electrical engineering.
What they built is called a molecular memristor — and it is unlike anything that has come before it.
📖 GLOSSARY: What is a Memristor?
A memristor (memory + resistor) is an electronic component that can change and remember its resistance. Unlike a standard resistor that always has the same resistance, a memristor's resistance depends on the history of current that has passed through it. In other words — it remembers. This makes it a natural candidate for mimicking synaptic behaviour in the brain.
The researchers synthesised 17 carefully designed ruthenium (RuII) complexes — molecules built around a central ruthenium atom, surrounded by specially engineered organic ring structures called azo-aromatic ligands. By precisely adjusting these ligands and the surrounding ionic environment, they discovered something extraordinary:
The same single molecular device could function as five completely different electronic components — depending only on how it was electrically stimulated:
| Function | What It Does | Brain Equivalent |
|---|---|---|
| Memory Element | Stores data persistently | Long-term memory |
| Logic Gate | Makes decisions (AND/OR/NOT operations) | Neural decision-making |
| Selector | Routes signals intelligently | Attention mechanism |
| Analog Processor | Handles continuously varying data | Continuous sensory processing |
| Electronic Synapse | Learns and adapts from experience | Synaptic plasticity — how learning works |
Professor Sreetosh Goswami described it simply: "It is rare to see adaptability at this level in electronic materials. Here, chemical design meets computation — not as an analogy, but as a working principle."
And his co-author, Visiting Scientist Sreebrata Goswami, who led the chemical design, said something that should be printed in every chemistry classroom in the country: "This work shows that chemistry can be an architect of computation, not just its supplier."
⚗️ The Chemistry Behind It: Why Ruthenium?
This is the part that makes me genuinely excited as a materials chemist — because the elegance of the chemical design is the entire story.
Ruthenium is a transition metal — element 44 on the periodic table — belonging to the platinum group. It is valued in chemistry for something very specific: its ability to exist in multiple oxidation states and to form highly stable coordination complexes with organic ligands. This makes it extraordinarily useful for systems where you need controlled, predictable, tunable electron behaviour.
The IISc team built their ruthenium complexes with azo-aromatic ligands — nitrogen-containing organic ring structures that are "non-innocent," meaning they can themselves accept or donate electrons. This creates a rich landscape of electronic states within a single molecule.
The key insight was this: by adjusting the specific ligands and the ionic environment surrounding the ruthenium centre, the researchers could tune exactly how electrons and ions flow through the device. Small, precise molecular-level changes produced completely different macroscopic electronic behaviours. This is what they call structure-function engineering — the deliberate design of a molecule's geometry to produce a specific functional outcome.
🔑 The Key Chemical Mechanism: The ruthenium complex uses what the researchers describe as "zeroth-order kinetics" for switching — meaning the molecular state transition happens at a constant, predictable rate regardless of external conditions. This predictability is what makes it possible to engineer specific electronic functions by design rather than by trial and error. It is the difference between guessing and knowing.
The device achieved impressive benchmarks: 14-bit analog resolution and an energy efficiency of 4.1 TOPS per watt (Tera Operations Per Second per watt). To put that in context — this means the device can perform an enormous number of computational operations while consuming very little energy. The human brain operates at roughly 10-20 TOPS per watt. Conventional AI accelerator chips typically achieve 1-5 TOPS per watt. This molecular device is already competitive.
The analog resolution deserves special mention. Standard digital electronics store information in binary — 1 or 0. A 14-bit analog device can represent 16,384 distinct levels of conductance. This is precisely how biological synapses work — not with on/off switches, but with continuously graded signal strengths that encode the weight of a memory or the strength of a learned association.
🧩 An Analogy That Makes This Concrete
Imagine you are building a city. In the conventional approach, you build separate buildings for each function: one building for storage, another for processing decisions, another for routing traffic. These buildings are connected by roads, and information must travel between them constantly. This is expensive, slow, and consumes enormous amounts of energy just in the transportation of information.
Now imagine you build a single structure that can be a library on Monday, a courthouse on Tuesday, a hospital on Wednesday, and a school on Thursday — depending on who walks through the door and what they need. No roads required. No energy lost in transportation. The building itself is intelligent enough to reconfigure.
That is what this ruthenium molecular device is. And because memory and processing happen in the same physical location — in the same molecule — there is no energy wasted moving data from one place to another. This is called in-memory computing, and it is one of the most important concepts in next-generation hardware design.
Your brain already does this. It has done it for millions of years. The IISc team has now found a way to replicate it in a designed molecule.
🇮🇳 Why This Is a Proud Moment for Indian Science
The global race to build post-silicon computing hardware is intensely competitive. Major research groups in the United States, Europe, China, Japan, and South Korea have been working on neuromorphic devices for decades. IBM, Intel, and Samsung have invested billions of dollars in this space.
This discovery did not come from a Silicon Valley lab or a European research institution. It came from IISc Bangalore — the city many of us live and work in — through what was described as a collaboration spanning chemistry, physics, and electrical engineering. It is a genuinely interdisciplinary achievement, and it is one that places India firmly at the frontier of molecular electronics and neuromorphic computing.
Professor Sreetosh Goswami has been awarded the 2025 iCANX Young Scientist Award and the Small Young Innovator Award for his contributions in nanoscience. His group is already working on translating this research into a prototype chip — to be built on a 22-nanometre process using a crossbar architecture — through a startup being incubated at CeNSE. A physical chip is expected within the next year.
This is the moment where academic discovery meets real-world engineering. And it is happening in India.
🌐 The Bigger Picture: What This Means for AI
The implications of this research extend far beyond a single paper.
Artificial intelligence as we know it is facing a crisis of energy consumption. Training a single large language model can consume more electricity than an average American home uses in a decade. As AI systems grow more powerful and more pervasive, the energy demands are becoming a genuine global concern. The world does not have enough power plants — let alone sustainable ones — to run the AI future we are building on current hardware principles.
Molecular neuromorphic devices represent a fundamentally different path. Rather than simulating intelligence in software running on power-hungry silicon, they physically embody intelligence in matter. They do not imitate a brain — they function according to the same principles a brain uses. And because the mechanism is intrinsic to the material's chemistry rather than engineered around its limitations, the potential for efficiency gains is enormous.
Beyond energy, consider the implications for edge computing — the idea of running AI directly on small devices rather than in massive data centres. Hearing aids that adapt to acoustic environments in real time. Medical implants that learn from physiological signals. Sensors in manufacturing environments that detect anomalies without needing a cloud connection. All of these become possible when the computing hardware is small, low-power, and genuinely adaptive — which is exactly what molecular memristors offer.
"Unlike conventional electronics, these devices do not just imitate intelligence. They physically encode it."
— IISc Press Release, December 2025
This is the distinction that matters. Current AI hardware simulates learning by repeatedly running mathematical operations across billions of transistors. These molecular devices are the learning — the intelligence is encoded in the chemistry of the material itself. That shift, from simulation to embodiment, is the most significant conceptual breakthrough in computing hardware in half a century.
🔭 What Comes Next
The IISc team is not stopping at proof-of-concept. They are actively working on several fronts simultaneously:
1. Silicon integration: The team is working to place these molecular materials directly onto conventional silicon chips. This is a critical step — it means the technology does not require a completely new manufacturing infrastructure. Molecular devices could potentially be layered onto existing chip fabrication processes, dramatically accelerating the path to commercial viability.
2. Crossbar architecture: The prototype chip will use a crossbar design — a grid of intersecting wires where each intersection point contains a molecular memristor. This architecture allows massive parallelism, similar to how the brain processes information across networks of neurons simultaneously rather than sequentially.
3. Startup development: A startup is being incubated at CeNSE to commercialise this technology. This is the bridge between laboratory discovery and real-world application — and it signals that the researchers themselves believe this is not merely an academic result but a genuinely deployable technology.
4. Expanding the molecular library: With 17 ruthenium complexes characterised, the team has established a systematic framework for predicting how molecular changes affect electronic behaviour. This predictive power — the ability to design a molecule and know what it will do before synthesising it — is the key to scaling this approach into a genuine technology platform.
💬 A Chemist's Perspective
I want to say something directly, as someone who works in R&D chemistry and whose own research touches on nanomaterials and molecular-scale phenomena.
For years, chemistry has been the quiet backbone of computing — providing the materials that engineers then shaped into devices. Chemists synthesised the silicon, purified it, doped it, and handed it to the engineers. The intelligence, the architecture, the function — that was always considered the domain of electrical engineering and computer science.
What this research represents is a fundamental inversion of that relationship. Here, the chemistry itself is the architecture. The molecular design determines the electronic function. A chemist who designs a ligand is now, simultaneously, programming a computing element. The boundary between chemical synthesis and circuit design has dissolved.
This is what excites me most about this discovery — not just what it means for AI, but what it means for what chemistry is and can be. We are at the beginning of an era where molecular design is computational design. Where a chemist's ability to engineer a coordination complex is indistinguishable from an engineer's ability to design a circuit.
If you are a chemistry student reading this, pay attention. The next generation of AI hardware is going to be built by people who understand molecules. That might be you.
📋 Quick Summary
✅ Who: Prof. Sreetosh Goswami's team at CeNSE, Indian Institute of Science, Bangalore
✅ What: A single molecular device (ruthenium complex) that can function as memory, logic gate, selector, analog processor, and artificial synapse
✅ How: Precise engineering of ligands and ionic environment around a RuII centre — electrons and ions reorganise dynamically in response to electrical stimulation
✅ Performance: 14-bit analog resolution, 4.1 TOPS/W energy efficiency
✅ Published: Advanced Materials, December 9, 2025 | DOI: 10.1002/adma.202509143
✅ Next steps: Startup incubation, silicon chip prototype on 22-nm process expected within a year
📚 References
1. Gaur P, Kundu B, Ghosh P, Bhattacharya S, et al. "Molecularly Engineered Memristors for Reconfigurable Neuromorphic Functionalities." Advanced Materials, 2025. DOI: 10.1002/adma.202509143
2. IISc Press Release: "Encoding adaptive intelligence in molecular matter by design." December 30, 2025. iisc.ac.in
3. ScienceDaily: "Beyond silicon: These shape-shifting molecules could be the future of AI hardware." January 3, 2026. sciencedaily.com
4. EE Times: "Indian Researchers Develop Molecular Memristor for Neuromorphic Computing." April 2026. eetimes.com
Written by Athira Vijayan
R&D Chemist | M.Sc. Applied Chemistry (2nd University Rank) | Nanomaterials Researcher
Working on carbon dot synthesis and functional nanomaterials in Bangalore, India.
📸 Instagram: @know_chemicals | 🔗 ORCID: 0009-0009-4646-0209 | ✉️ athiravjn99@gmail.com
No comments:
Post a Comment