Ternary computing and memristors: going beyond binary logic
Memristors open the door to neuromorphic computing, which mimics the operation of biological synapses and has the potential to significantly accelerate the development of artificial intelligence. A transition to ternary computing, combined with memristor-based architectures, could become one of the paths toward overcoming the current limitations of classical binary logic, offering substantially greater energy efficiency and computational performance for future computing systems.
For over 70 years, the world of computing has relied on one simple idea: about everything is either a 0 or 1. That foundation has given us smartphones, tablets, computers, the internet, and now, artificial intelligence.
But here’s the catch: data centers are consuming enormous amounts of electricity just moving data around. And also, it’s expensive.
The researchers are now looking beyond binary logic in search of something more efficient and affordable. They’re pointing to the same concept: ternary logic for computers to work like brains, not calculators.
If computing is about to experience another shift, this could be it.
What is binary computing?
The world we know is using binary computing, which represents all information with only two states, a 0 and 1. That model has powered about everything, from websites to satellites.
For decades, this worked extremely smoothly because transistors are excellent at switching between 0s and 1s. But today, we’re running into barriers.
What is ternary computing?
That’s when the idea of using ternary computing, which suggests simply adding a -1, another state, comes in. That shift might seem quite small at first but changes the picture a lot.
This allows one element to store more information while performing fewer operations and using less energy. Many argue, this approach is even more “natural”.
A shift in hardware
What is a memristor?
A memristor (or simply memory resistor) can change its resistance simply based on previous electrical charge. In brief, it “remembers” previous states when power’s cut off.
Fun fact: it was predicted mathematically in 1971 and called a fundamental (basically fourth) circuit element.
Just imagine an adaptive water pipeline that changes its diameter in dependence on water flowing through. Now imagine, it keeps that diameter when water suddenly stops.
Cool, right?
How is that relevant?
That’s exactly where memristors become interesting – they’re breaking the limits of just two states, the 0 and 1. That means, a memristor naturally fits ternary logic.
In brief: a memristor is the irreplaceable hardware that can turn balanced ternary computing into reality.
Why is this important?
Data centers already consume enormous amounts of electricity
In 2024, the global data centers have consumed 415 TWh of power, roughly 1.5% of worldwide electricity use. By 2030, the demand might rise and reach 945 TWh per year, which is about 3% of worldwide electricity use (and comparable to Japan’s electricity consumption).
Frontier models are becoming absurdly expensive in training
Researchers estimated that cost of training has been steadily growing at roughly 2.4x annually since 2016. Related expenses already account for tens of millions, and some runs could even exceed $1 billion by 2027.
Neuromorphic engineering – the next big thing?
AI technology is powerful but carries a very old habit: it moves data back and forth, and that’s very inefficient. AI seems very impressive, but only on the surface level: the memory, the learning, and then, the computing – that’s plenty of resources.
Our brain doesn’t split a task into several smaller steps, and that is why it can do much while using little power. And that’s the solution.
The researchers are working on creating a computer that remembers, can learn, and handles the processing. All simultaneously, without treating those separately: an idea that’s simple but ambitious and important.
That’s where a memristor is becoming very interesting – it’s similar to synapses (the space between neurons). Put simply, a memristor can soften the boundaries of memory and computation – the shift this field is chasing.
Neuromorphic computers are entering the stage
Neuromorphic computers are moving from theory into reality, and that’s what makes this topic so compelling. Intel, IBM, SynSense, BrainChip, Samsung Electronics, and laboratories across the United States, China, Japan, and Europe are working on systems that mimic the brain.
For now, most processors still rely on classical CMOS transistors – the idea is real, but there’s a lot to unravel.
Neuromorphic computers are believed to become the reality when memristors will break out of the laboratory. That could make computing more adaptive and way more efficient.
And that’s a whole new era we’re anticipating, not just an upgrade.
How humanity can benefit
A lower electricity consumption
Less heat, less cooling, less power to move data around, less waste.
No running on empty
Less strain, fewer bottlenecks, no more billion-dollar upgrades, fewer walls to hit.
Smarter devices
Phones, glasses, vehicles, drones, and other consumer devices – more capable.
Advanced robotics
More practical, useful machines, not just science projects (without compromising).
A less concentrated landscape
More players, fewer giants that dominate the market.
A sustainable path forward
More growth, less pressure on infrastructure and grids.
But is that possible?
Memristive synapses to cut the consumption of energy by 100x
A paper has shown memristive synapses could use two orders of magnitude less energy than pure GPU usage. The paper also says memory traffic (not computing) has wasted over 98% of energy in said GPU installation – and that’s why using in-memory operation is critical.
Memristive hardware to cut the consumption of energy and area
A paper on memristors has demonstrated a 15.1x increased energy efficiency and 12.9x reduced area usage. And more, a memristor has shown a 57.2% reduced energy overhead and 30.7% reduced area overhead when powering compute-in-memory systems.
How we can help
Ternary computers and memristors are emerging, but the forces behind the revolution are already right here. Mature organizations shouldn’t wait for downfall to fix what’s broken.
Let’s talk?
FAQ
A memristor is an electronic component that can basically remember how electricity flowed through it before. It’s basically a new circuit element that’s valued for memory.
The memristor is mainly being explored to support several areas:
- In-Memory computing
- Neuromorphic computing
- RRAM applications
- And security
A memristor is changing its resistance all based on amount and direction of charge that passed through before. A positive voltage moves tiny particles into the insulating layer, thus forming conductive pathways or filaments; a reversed voltage pulls those back from the insulating layer and breaks said pathways.
A memristor, when past the research, can power many things:
- Energy-efficient infrastructure
- Self-learning systems
- Smart devices
- Advanced robots
- Autonomous vehicles
- Advanced sensors, and more
At the very moment, memristor computing is still in its early-stage research and capitalization, so no, not yet. It’s expected to grow in line with demand.
To put it short, memristor models do exist in labs and prototypes, but aren’t widely deployed in electronics. The issue is that current hardware still relies on conventional CMOS-based technology.
Memristor memory is a non-volatile memory built around circuit elements that remember previous states. Ternary memory, if it becomes accessible, could blur boundaries between data storage and processing, thereby making computing systems far more energy efficient.
The difference comes down to how many states the system can represent: two, three, or even four states. More states mean that more information can be both stored and processed in one single element, that’s it.
Ternary computing systems represent more information, which means they store and process more of it too. Lower waste (electricity particularly), lower expenses, smarter devices, advanced robotics becoming accessible – the benefits are looking quite promising for humanity.
Ternary computing hardware isn’t that easy to produce and remains mostly confined to labs and prototypes. The promise is real, but delivering large-scale systems is still a major engineering challenge.