Making quantum computing components in a lab is one thing, but getting them to work using standard CMOS processes is another. Now, a team of researchers have successfully demonstrated the production of p-bits using standard CMOS technology. What challenges does the manufacture of quantum systems involve, what did the researchers develop, and how does this change the quantum game?
The Challenge of Building Quantum and Probabilistic Hardware
Conventional computing is built on one simple assumption: bits are deterministic. Every bit is either a 0 or a 1, and that state does not change unless it is explicitly updated, and it is this predictability that lays the foundation of all modern digital systems.
Quantum and probabilistic computing, however, breaks that assumption in just about every conceivable way.
Instead of fixed states, these systems rely on behavior that is inherently stochastic, fluctuating, or probabilistic by design. In principle, this opens up new ways of solving problems, particularly in optimization and systems with many possible valid solutions. However, in practice, it introduces a very serious engineering problem that still remains a major hurdle.
Modern semiconductor manufacturing is not designed with uncertainty in mind, but instead, is built around repeatability, control, and tightly defined switching behavior. Every transistor is expected to behave the same way every time, within extremely tight tolerances. Quantum systems, however, need to push this uncertainty requirement to its absolute limit. Many quantum experimental platforms require cryogenic temperatures, specialized materials, and carefully engineered spintronic structures. Even the smallest deviations in temperature, material thickness, or fabrication quality can disrupt the fragile behavior required for quantum or probabilistic operation. And this is not just a matter of scale, as even small variations in CMOS fabrication, such as 100 nm layers varying between 90 nm and 110 nm, can introduce uncontrolled noise or completely destroy the intended behavior.
On top of this, standard CMOS processes are explicitly designed to suppress randomness. Any stochastic behavior is normally treated as noise and engineered out of the system. Turning that same “noise” into a useful computational feature is not something traditional processes were ever intended to support.
As a result, most early probabilistic hardware has relied on hybrid systems, external noise sources, or laboratory-only setups that do not scale well into integrated circuits.
Researchers Create the First CMOS-Compatible Probabilistic Bit
Recently, a joint research team from Japan and the United States has demonstrated what is reported to be the world’s first integrated spintronic probabilistic bit (p-bit) fabricated directly using a standard CMOS process. The work was published in IEEE Electron Device Letters and represents a significant step toward practical probabilistic computing systems, sometimes referred to as “p-computers”.
Unlike conventional bits, which are fixed at either 0 or 1, a p-bit continuously fluctuates between both states over time (i.e. a superposition state). This makes p-bits useful for problems such as optimization, where exploring many possible states is often more valuable than evaluating a single deterministic solution.
The implementation developed by the researchers combines standard CMOS circuitry with spintronic devices. CMOS transistors were fabricated using a 130 nm process provided by SkyWater Technology, while the spintronic superparamagnetic nanodevices and electrodes were produced using facilities at Tohoku University.
For the researchers device, spintronics plays a crucial role as nanoscale magnetic systems naturally exhibit stochastic switching behavior. At very small scales, magnetic orientation can fluctuate due to thermal effects, producing controlled randomness that can be harnessed rather than suppressed. Thus, in this design, the key idea was not to eliminate randomness, but to control it.
Instead of relying on random current fluctuations, the system uses magnetic state changes to generate probabilistic switching with lower energy cost and higher stability. The researchers confirmed that the device meets two essential requirements for a functional p-bit: random temporal switching between states and controllable average output using an input signal.
This is also the first demonstration of a spintronic p-bit integrated directly into a CMOS-compatible workflow, rather than being assembled as a standalone laboratory device.
While this does not immediately lead to large-scale systems, it does provide a path toward higher-density probabilistic circuits built using existing semiconductor infrastructure.
Could CMOS P-Bits Change the Future of Computing?
The key point about this development is not that p-bits replace traditional transistors, but that they extend what standard silicon can do.
Probabilistic computing is most useful in areas where uncertainty is part of the problem space. This includes optimization problems, machine learning inference, and combinatorial search tasks, where exploring many potential solutions is more efficient than calculating a single deterministic result. By embedding probabilistic behavior directly into hardware, these systems could reduce the computational cost of certain classes of problems that are inefficient on classical architectures.
An additional advantage is practicality. Unlike quantum computing approaches that require extreme operating conditions, CMOS-based p-bits operate at room temperature and can, in principle, be manufactured using existing semiconductor infrastructure. This makes them far more realistic in the near term, even if their capabilities are more limited.
However, significant challenges do still remain, making this device far from ideal. Noise control, device variability, and large-scale integration are still open engineering problems. These issues also mean that it is unlikely for p-bit systems to ever replace CPUs or GPUs, since deterministic logic remains far more efficient for general-purpose computation.
Instead, the more realistic outcome is that they act as specialized accelerators within hybrid systems, handling specific probabilistic workloads while conventional silicon continues to manage everything else.
You may also like
