CMOS-integrated superparamagnetic tunnel junction-based p-bit
Juyoung Yoon, Nuno Caçoilo, Advait Madhavan, Jabez J. McClelland, Shun Kanai, Hideo Ohno, Shunsuke Fukami, and William A. Borders · 2026
Read CMOS-integrated superparamagnetic tunnel junction-based p-bit by Juyoung Yoon, Nuno Caçoilo, Advait Madhavan, Jabez J. McClelland, Shun Kanai, Hideo Ohno, Shunsuke Fukami, and William A. Borders (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-14446.txt.
Introduction
Yoon and seven colleagues demonstrate a probabilistic bit unit cell that pairs superparamagnetic tunnel junctions with 130-nanometer CMOS technology. The junction's resistance fluctuations drive a matching fluctuating digital output voltage, tunable through the input voltage. Crossing the CMOS-integration threshold has blocked scaling of junction-based probabilistic systems, and this cell shows such integration working. The result establishes feasible CMOS-compatible probabilistic circuits and a step toward scalable hardware for hard optimization and learning problems.
Catalog record
| CREATOR | Juyoung Yoon, Nuno Caçoilo, Advait Madhavan, Jabez J. McClelland, Shun Kanai, Hideo Ohno, Shunsuke Fukami, and William A. Borders |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 12822 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.14446
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
CMOS-integrated superparamagnetic tunnel junction-based p-bit Juyoung Yoon, Nuno Caçoilo, Advait Madhavan, Jabez J. McClelland, Shun Kanai, Hideo Ohno, Shunsuke Fukami, and William A. Borders Abstract Probabilistic computers offer promising solutions for computationally hard problems in domains such as combinatorial optimization and machine learning. A key building block in these systems is the probabilistic bit (p-bit), which relies on superparamagnetic tunnel junctions (sMTJs) as its source of randomness. A challenging threshold to cross for scaling sMTJ-based p-bit systems is integration of sMTJs with CMOS technology. In this work, we present experimental results of a p-bit unit cell using sMTJs integrated with 130 nm CMOS technology and demonstrate that the sMTJ’s resistance fluctuations can generate a corresponding fluctuating digital output voltage which is tunable via the input voltage. These findings establish the feasibility of CMOS-compatible, sMTJ-based probabilistic circuits and mark a key step toward scalable hardware for real-world probabilistic computing applications. I Introduction Computationally hard tasks like combinatorial optimization and machine learning are deemed difficult for conventional deterministic computers. Probabilistic computing is an alternative computing method using probabilistic bits (p-bits) fluctuating in time between a digital high and low state [ 1 ] . When computationally hard problems are formulated into an Ising Hamiltonian [ 2 ] , these fluctuations allow the computer to probabilistically sample a multitude of configurations. Recent probabilistic computers using field-programmable gate arrays (FPGAs) have demonstrated increasingly larger problem sizes [ 3 ] by utilizing concepts like parallel-tempering [ 4 ] and sparse encoding [ 5 ] . To tackle harder problems, forecasts state that the most efficient p-computers will use unstable nanomagnets as the source of randomness [ 1 ] . A superparamagnetic tunnel junction (sMTJ) is one such unstable nanomagnet with potential for providing the randomness in a p-bit. sMTJs are composed of ferromagnetic free and reference layers separated by a tunnel barrier. The sMTJ can show a high or low resistance, depending on the angle of orientation between the two ferromagnetic layers owing to the tunneling magnetoresistance (TMR) effect [ 6 ] . In contrast to magnetic tunnel junctions (MTJs) used in magnetoresistive random access memory [ 7 ] [ 8 ] , sMTJs feature drastically reduced energy barriers so thermal fluctuations cause the state of the device to probabilistically change on a nanosecond timescale [ 9 ] , [ 10 ] . So far, sMTJ-based p-bits have been used to solve a multitude of problems [ 11 ] - [ 16 ] and are typically represented with a unit cell circuit consisting of an sMTJ in series with an n-channel MOS (NMOS) transistor whose drain output is fed into an inverter that produces binary voltage fluctuations between high and low [ 17 ] . The NMOS gate voltage is used to change the current flowing through the sMTJ which tunes the time the sMTJ spends in one state [ 11 ] [ 12 ] . Current p-bit prototypes typically consist of sMTJs wire-bonded to a circuit board with an NMOS and an inverter; however, to show their full potential for practical applications, results showing operation on a large-scale ASIC are required. Therefore, a key phase of sMTJ-based p-bit development is the systematic study of back-end-of-the-line (BEOL) integration of s-MTJs with CMOS. This involves isolating each stage of the p-bit circuit to quantitatively evaluate properties suitable for both the CMOS and the sMTJ. In this work, we successfully integrate sMTJs with 130 nm CMOS technology, evaluate challenges of integration at each stage of the circuit, and demonstrate the operation of a complete p-bit unit cell. II Chip Design and Fabrication We systematically design a CMOS-integrated p-bit circuit to verify its functionality at each maturity stage. The 5.2 mm × 3.6 mm monolithic chip containing 150 isolated sMTJs, 240 sMTJs in series with an NMOS transistor, and 150 full-stage p-bit circuits, is fabricated using a 130-nm commercial CMOS process line and a dedicated MTJ process line at Tohoku University (Fig. 1 a,b). We design the sMTJ + NMOS circuit with electrodes for the gate, source, power-supply voltage \textit{V}_{\text{DD}} , and output connections. For the full-stage p-bit circuit, we design a cascode stage in series with the sMTJ to reduce the effect that the fluctuating resistance of the sMTJ has on the modulation of the source-drain current. The output feeds to the input of a variable threshold controller (VTC), containing two pull-up and two pull-down transistors, that collectively represent threshold voltages from 0.7 V to 1.1 V with a 100-mV step. The VTC stage output feeds into a final inverter which produces voltage fluctuations between 0 V and 1.8 V. The circuits for the sMTJ + NMOS and full stage are completed by patterning an sMTJ between the output (or input to VTC) pads containing vias to the CMOS and \textit{V}_{\text{DD}} . To enable a wide range of currents, the transistors are designed with channel widths of 1, 3, 9, 27 µm, producing currents from hundreds of µA to tens of mA. To enable BEOL integration of sMTJs, the final encapsulating CMOS metal layer is left off which gives direct contact to metal and tungsten via layers below. Leaving the final metal layer off helps keep the roughness of the BEOL interface to a minimum, but carries the risk of the tungsten vias oxidizing. To prevent this, wafers were immediately coated in photoresist, where a mild removal process was used to clean the surface before material stack deposition. After photoresist removal, we first performed four-point probe measurements on the top and bottom electrodes to confirm ohmic contact. We then deposit the sMTJ stack above the CMOS circuits by DC and RF magnetron sputtering at room temperature of the following stack structure: Ta(5)/ PtMn(20)/ Co(2.4)/ Ru(0.9)/ CoFeB(2)/ MgO/ CoFeB(2.1)/ Ta(5)/ Ru(5)/ Ta(50) (thicknesses in nm). The sMTJs are patterned into an elliptic shape with effective diameters of 50, 60, 70, 80 nm and aspect ratios ranging from 1, 2, 3, 4 by Ar+ ion milling and reactive ion etching (Fig. 1 c,d). Figure 1: (a) Layout schematic of test chip with 150 isolated sMTJs, 240 sMTJs in series with an NMOS transistor, and 150 full-stage p-bit circuits. b) Schematic of the cross-sectional structure of the integrated chip. c) cross-sectional scanning transmission electron micrograph of an sMTJ integrated above the CMOS. d) Plan-view scanning electron micrograph of an sMTJ. Figure 2: Time-averaged resistance versus applied current for a single sMTJ with 25 nm × 100 nm dimensions. (inset) Experimental measurement setup. III Experimental Results To confirm electrically tunable stochastic behavior of the isolated sMTJs, the time-averaged resistance <\textit{R}> is evaluated as a function of applied current in a single sMTJ (Fig. 2 ). Measurements were performed by connecting DC probes to top and bottom electrodes, applying a DC current, and reading voltage fluctuations on an oscilloscope. The time-averaged voltage at each current is read through the same source-measure unit supplying the current and averaged for 100 samples taken at 10 ms intervals. As the current sweeps from -80 µA to 50 µA, the sMTJ changes from spending more time in the low resistance state to spending more time in the high resistance state. With the varied diameters and aspect ratios, we measured a wide range of low resistance values