Scientists created flexible probabilistic bits from custom polymers, offering a new, energy-efficient path for AI and machine learning using classical physics.
A team of scientists has built a new type of electronic element that could pave the way for smarter, more efficient computers. Using a custom-designed polymer, the researchers have shown that computing units known as probabilistic bits can be created from soft, carbon-based materials. These flexible components behave differently from traditional electronics and could offer a new path toward energy-saving machines designed to tackle problems that stretch the limits of current technology.
“Probabilistic bits are binary units that randomly fluctuate between 0 and 1 with a tunable probability,” Stephen Foulger, a materials scientist at Clemson University and the study’s lead author, said in an email. “Unlike classical bits which are fixed in state, p-bits incorporate controlled randomness governed by classical (not quantum) laws. This makes them ideal for multiple applications such as artificial intelligence and machine learning systems.”
The study, published in Advanced Physics Research, explores the behavior of a material known as pTPADTP — a polymer made from carbon, hydrogen, nitrogen, oxygen, and sulfur. Polymers are large molecules formed by repeating smaller building blocks, and they appear in a wide variety of everyday materials, from rubber and plastics to DNA. The particular polymer used here was engineered with chemical features that allow electric charges to move in unpredictable ways — a key ingredient for generating probabilistic behavior.
Room-temperature computing with built-in randomness
Although p-bits may sound like an attempt to mimic the qubits of quantum computing, the resemblance is only superficial. The underlying physics is very different. While qubits derive their probabilistic nature from the bizarre laws of quantum mechanics, including superposition and entanglement, p-bits rely entirely on classical effects such as thermal noise and resistance fluctuations.
“P-bits can be realized using classical hardware at room temperature, while q-bits require cryogenic and quantum-coherent environments,” said Foulger.
That makes them far easier to fabricate and more compatible with the kinds of chips and systems already used in today’s computing infrastructure. While they may not offer the same power as fully quantum processors for tasks like simulating molecules, their simplicity and versatility make them highly attractive for specific kinds of calculations, particularly those involving probability, noise, and uncertainty.
Polymers as a platform for smart hardware
The new work builds on Foulger’s group’s previous experience in creating electronic components from various kinds of polymers. These long-chain molecules can conduct electricity and be tuned through chemical modifications to behave in useful ways. For this study, the researchers focused on achieving controlled randomness through molecular design.
In their new polymeric system, the p-bits switch randomly between two different resistance levels, which represent the binary values 0 and 1. By carefully tuning how likely the material is to switch between high and low resistance, researchers can control the behavior of the p-bits — a key feature for building energy-efficient computing systems that rely on randomness.
The device’s behavior was validated experimentally against theoretical models of stochastic neurons and showed strong agreement. The team also demonstrated that the probability of the bit being in the “1” state could be finely adjusted using small voltage offsets.
A milestone in probabilistic electronics
Foulger noted that while earlier work in this area was spearheaded by Professor Supriyo Datta and colleagues at Purdue University, who used magnetic tunnel junctions as the basis for their designs, this new approach relies on dynamic rearrangements in the polymer’s molecular structure to produce a similar effect.
This organic alternative offers a major advantage: flexibility in processing and material selection, along with low-temperature manufacturing. These properties make it particularly suitable for neuromorphic systems — electronic devices inspired by how the brain processes information — and for machines that need to carry out large-scale probabilistic computations using little energy.
The applications for p-bits are wide-ranging. According to Foulger, “They can accelerate machine learning tasks and solve combinatorial optimization problems. Their intrinsic randomness also makes them ideal for secure computation, including true random number generation and physically unclonable functions.”
Looking ahead
Potential applications for p-bits are broad. They include solving optimization problems such as finding the shortest route through a network, speeding up certain machine learning methods, generating random numbers for security, and building components that are hard to copy or fake.
Beyond computation, Foulger’s group is working on two related challenges. The first is uncovering how a polymer’s internal structure shapes its unpredictable behavior — knowledge that could help engineers design even more efficient materials. The second is building full-scale hardware systems that make use of many p-bits working together in parallel, something that would bring the technology closer to practical deployment.
With this achievement, the researchers have shown that soft, carbon-based materials can do more than just carry current — they can support entirely new forms of information processing. By capturing randomness in a controlled way, p-bits built from polymers could help fill the gap between today’s rigid computing architectures and the probabilistic logic of tomorrow.
Reference: Stephen H. Foulger et al, Polymeric Memristors as Entropy Sources for Probabilistic Bit Generation, Advanced Physics Research (2025). DOI: 10.1002/apxr.202400142.
Feature image credit: geralt on Pixabay
