Abstract
Information security and computing, two critical technological challenges for post-digital computation, pose opposing requirements – security (encryption) requires a source of unpredictability, while computing generally requires predictability. Each of these contrasting requirements presently necessitates distinct conventional Si-based hardware units with power-hungry overheads. This work demonstrates Cu0.3Te0.7/HfO2 (‘CuTeHO’) ion-migration-driven memristors that satisfy the contrasting requirements. Under specific operating biases, CuTeHO memristors generate truly random and physically unclonable functions, while under other biases, they perform universal Boolean logic. Using these computing primitives, this work experimentally demonstrates a single system that performs cryptographic key generation, universal Boolean logic operations, and encryption/decryption. Circuit-based calculations reveal the energy and latency advantages of the CuTeHO memristors in these operations. This work illustrates the functional flexibility of memristors in implementing operations with varying component-level requirements.
Similar content being viewed by others
Introduction
The big data era addressed by artificial intelligence (AI) brings two critical issues—information security and the energy cost of the associated hardwares1,2. Memristors, or two-terminal nonvolatile memories, have been extensively studied as energy-efficient solutions for both security and computing challenges in the post-digital era of computing. Security solutions using memristors, such as the generation of true random numbers and physically unclonable function (PUF), leverage the stochastic randomness in the memristors during the switching processes3,4,5,6. On the other hand, memristor-based systems used for low-power in-memory computing (wherein computing and memory units are combined) strive to achieve a low degree of component-level randomness and variability7,8,9. Despite these advances, these solutions have been built using different types of memristors, often on dedicated hardware units, which limit their generality. Besides, even when memristor hardware is designed for in-memory computing, the inherently higher variability of memristors renders the hardware suitable for probabilistic computing (such as matrix multiplications) but not for highly deterministic operations (such as arithmetic and Boolean logic). This issue further limits the generality of memristor-based computing. Thus, despite the high energy efficiency of memristor-based computing systems demonstrated for specific applications, they still fail to compete with highly flexible digital processors.
This work demonstrates a highly functional scheme of memristor-based hardware that integrates security, computing and memory capabilities using the Cu0.3Te0.7/HfO2 (CuTeHO) ion-migration-driven memristor. Its stochastic switching variation is exploited as an entropy source for generating PUFs, with the added features of concealability and reconfigurability. When operated away from biases that promote stochasticity, a single CuTeHO memristor device can be guided via a sequence of up to three switching operations to exhibit all Boolean logic operations. Furthermore, this work demonstrates multiple computing processes performed on memristive hardware by exploiting the dual functional nature of the CuTeHO memristors: unclonable key generation, logic operations, and encryption/decryption, thereby demonstrating functional flexibility that has been missing in memristor-based systems. The energy consumption and latency of these operations are notably lower compared to prevailing digital computing approaches. While there are also performance advantages (compared to prior works that used memristors) in both PUF generation and Boolean logic, the novelty of this work is the functional ability to switch between them on demand. Such flexibility enables an essential step towards functionally dense post-digital computers. As a demonstration of the utility and effectiveness of the encryption-decryption scheme, using crossbar arrays constructed using different nonvolatile memristors, matrix multiplication operations are performed on encrypted data without a noticeable loss of information.
Results
Influence of CuxTe1-x compositions on the resistive switching behavior
The CuxTe1-xHO memristors have different switching behaviors depending on the Cu concentration (x) in the CuxTe1-x electrode. It exhibits volatile threshold switching (TS) behavior at x = 0.1 (Cu0.1Te0.9/HfO2, Fig. 1a, Supplementary Note 1 and Supplementary Fig. 1)10,11,12,13,14, and nonvolatile resistive switching (RS) behavior at x = 0.3 and 0.6. Figure 1b displays the current–voltage (I–V) curves of the memristor with x = 0.3 (Cu0.3Te0.7/HfO2, referred to henceforth as CuTeHO) memristor under a compliance current (Icc) of 10 nA, exhibiting a self-rectifying behavior. In contrast, the CuxTe1-xHO memristor with x = 0.6 (Cu0.6Te0.4/HfO2), which forms even stronger Cu conductive filaments (CFs), exhibits conductive bridge random access memory (CBRAM) behavior with Ohmic conduction in a low resistance state (LRS) (Fig. 1c and Supplementary Fig. 2a). Previous studies reported that the filament’s stability is strongly influenced by its surface curvature10,11,12,13,14, which determines the device to be either a diffusive memristor or a CBRAM. However, at x = 0.3, the device exhibits nonvolatile and rectifying behavior, which deviates from the typical characteristics of CBRAM. It exhibits an Ohmic conduction property immediately after being switched to the LRS by the I–V sweep (Curve 1 in Fig. 1b). However, the subsequent sweep does not exhibit Ohmic conduction, suggesting the potential coexistence of another switching mechanism (Curve 2). The conduction behavior observed in the LRS and the rectification characteristics in the negative bias region (Curve 3) are similar to that of an electronic switching memristor, where electron trapping and detrapping occur at a trap layer15,16,17,18. In addition, the I–V characteristics of the CuTeHO memristor were examined at a higher Icc of 100 nA (Supplementary Fig. 2b), where sufficiently strong CFs form, which cannot be ruptured at zero bias. Similar to the Cu0.6Te0.4/HfO2 device, no rectifying behavior was found due to the stronger CF at higher Icc. Furthermore, the I–V curves of the LRS device were measured at different temperatures (Supplementary Fig. 3), where the absence of a decrease in the current in the LRS device with increasing temperature indicates that the conduction is not solely based on a metallic CF. Based on the varying qualitative memristive behaviors, the switching mechanism of the CuTeHO memristor can be understood as a combination of CF formation/rupture and electron trapping/detrapping, as illustrated in Fig. 1d. The relatively weak Cu CFs are formed during the SET (switching from high resistance state (HRS) to LRS) process. At zero bias, the diffusion of the Cu atoms from the CF to the surrounding HfO2 region causes a slight rupture of the filament (similar to the behavior observed in the Cu0.1Te0.9/HfO2 device), simultaneously creating a localized path for charge trapping. In other words, the residual Cu CFs near the bottom electrode (BE) can be considered an extended electrode. The electron trapping/detrapping process in the local CF-ruptured region mediated by traps at roughly 0.7 eV below the conduction band (Fig. 1e), reported elsewhere18, can explain the observed switching behaviors. Kim et al.19 confirmed this switching mechanism, where the insulator was divided into switching and filament regions. The self-rectifying behavior in the negative bias region is attributed to the slightly ruptured filament and a possible barrier formed at CuxTe1-x/HfO2 interface due to the Fermi level pinning.
The self-rectification of the CuTeHO memristor can effectively suppress sneak currents when integrated into a passive memristive crossbar array, making it highly attractive for applications requiring low power consumption and high density. To further validate the practical use of the device in a large-scale array, using circuit-based simulations, sneak currents and line resistance issues were evaluated as the array size was increased (Supplementary Note 2). With the sneak current suppressed by the rectifying behavior, no critical issues were observed in array sizes up to 400 × 400 in the worst-case scenario and 1000 × 1000 when half the devices were at HRS (Supplementary Fig. 4). The line resistance did not affect the operation in a large-scale array (Supplementary Fig. 5). Additional details about the device are provided in Supplementary Figs. 6 and 7. The volatile and nonvolatile behaviors for the cases of x = 0.1 and x = 0.6 are not surprising as reported elsewhere20,21,22,23. However, the dual behavior at an intermediate stoichiometry (x = 0.3) is unexplored and is utilized for diverse applications in this work, which are explained in the following sections.
Physically unclonable functions with concealability and reconfigurability
PUF is a hardware security primitive that utilizes the inherent variations arising from individual components, such as manufacturing variations, cycle-to-cycle noise, and chaos, thereby providing each hardware with distinct properties and fingerprints24. PUF can be used for identity authentication and generating cryptographic keys for encrypting confidential or private data. The encryption process using the CuTeHO memristor is described in more detail later. Each PUF is associated with a unique output (‘response’) mapping from an input (‘challenge’), forming a challenge-response pair. Various PUF systems have been reported based on optical components, arbiters, static random-access memory, and ring oscillators20,21,22,23. However, the previously reported PUFs faced limitations with regard to scalability, fabrication complexity, non-trivial operation, and post-processing issues.
In contrast, the intrinsic stochasticity of memristors during their switching process holds great promise for hardware security applications owing to their high switching speed, low power consumption, and remarkable potential for scalability25,26,27,28,29,30. A critical requirement for modern PUFs is reconfigurability, where keys (the challenge-response pairs) are regenerated within the same chip to account for the increasing probability of information leakage with time. A second critical requirement is the concealability of the key when stored in hardware. There have been reports on reconfigurable memristive PUFs26,27,28,29, for instance, using conductance variations of neighboring cells to achieve reconfigurability. However, these reconfigurable memristive PUFs were not area-efficient as they required at least two memristors to generate a single bit. Furthermore, there has been a report of the concealment of keys in a memristive PUF, where the data was concealed by switching all devices into the LRS30. However, no memristive PUF system has offered both concealability and reconfigurability.
This work introduces a novel memristive PUF hardware that is concealable and reconfigurable, using the stochastic variations in the SET voltage of the CuTeHO memristor. A distribution of SET voltages from 400 cycles in 8 memristors was obtained using the cycle-to-cycle and device-to-device variations, with a median bias of 2.85 V used as the challenge that applies to all memristors for generating the PUF (Fig. 2a). The bit response is represented as ‘1’ when the memristor switches to the LRS and ‘0’ when it remains in HRS. Thus, this PUF hardware is area-efficient as each memristor represents one bit.
The experimentally fabricated 4 × 4 crossbar composed of the CuTeHO memristors was employed to demonstrate the memristive PUFs (Fig. 2a, b), of which fabrication procedure is provided in Supplementary Fig. 8. Concealing-revealing and reconfiguration processes are illustrated in Fig. 2c, d. The median value (2.85 V) was initially applied to all the memristors. This operation rendered half of them switched to LRS and the other left at HRS, due to the stochasticity of the CuTeHO memristors, generating PUF-1 (or base PUF) displayed in Fig. 2b. Subsequently, an additional I–V sweep down to −3.5 V, which partially ruptures the CFs in the LRS cell (step 2 of Fig. 2c), was applied to all the devices. This operation concealed the original PUF data because all the cells switched to HRS (step 3 of Fig. 2c). During concealment, the CFs rupture to different extents depending on the original state (LRS ruptures partially, while HRS ruptures fully; see Supplementary Fig. 9). Upon applying +2.85 V, the partially ruptured CFs were restored to the LRS, while the fully ruptured ones remained in the HRS, thus revealing the original key (step 4 of Fig. 2c). In this way, PUF-1 was concealed and revealed. When a key is discarded and a new key is desired (reconfiguration), all the memristors are made to undergo full SET (3.5 V) and full RESET (−4.0 V) cycles, which are followed by another I–V sweep to 2.85 V, as displayed in Fig. 2d. Once the device is programmed to its HRS, the SET voltage randomly changes after the RESET process (Supplementary Fig. 10). In other words, there is no history-dependent behavior that can compromise the randomness of a newly generated key. Hence, the PUF can be reconfigured with a different pattern of LRS/HRS cells due to the inherent stochasticity.
By including two consecutive reconfigurations of PUF-1, 20 different 4 × 4 PUFs were produced (the 16 × 20 bitmap in Fig. 2e and Supplementary Fig. 11). From these PUFs, the inter-Hamming distance (inter-HD) was calculated by comparing the bit difference between two PUFs. The histogram of the inter-HD was fitted with a Gaussian distribution with a mean value of 0.5063, showing excellent uniqueness of the memristive PUFs (Fig. 2f). In addition, all the collected bits in this work passed the National Institute of Standards and Technology (NIST) randomness test (Supplementary Table 1). Moreover, the conceal-reveal performance was examined through 10 consecutive cycles starting at PUF-1, achieving bit error rates of 1.88% for concealing and 0% for revealing (Fig. 2g and Supplementary Fig. 12).
Table 1 provides a comparison with other memristive PUFs. Among the qualitative features of reconfigurability, concealability, nonvolatility and logic capability, all prior efforts were unable to demonstrate at least two of them. This work provides all four qualitative features, in addition to area efficiency (a single cell represents a full bit). The device’s self-rectifying behavior also enables its implementation in selectorless (passive) crossbar array configuration, whereas prior works required complex design using transistors or selectors in order to be included in scalable crossbar arrays. The logic capabilities of the CuTeHO memristors are elaborated in the following section.
Dual conditional logic operations
Various methods of memristor-based in-memory computing have been proposed that can be categorized into stateful and nonstateful logic. In stateful logic, both logic input and output are represented by an identical information carrier, such as a memristor’s resistance7,31,32,33,34. On the other hand, nonstateful logic adopts different information carriers, such as voltage input and resistance output35,36,37,38. Stateful logic usually requires more memristors than nonstateful logic, but the logic cascading is more straightforward. In contrast, reconfiguring the nonstateful logic is more accessible than reconfiguring stateful logic since the logic operation can be adjusted by varying the input voltages. Because the CuTeHO memristor in this work possesses reconfigurability, a nonstateful logic approach is employed, and all 16 basic Boolean logic operations are demonstrated using a single device. This logic scheme has two voltage conditions, termed dual conditional logic (DC logic).
The fundamental DC logic operation method using the CuTeHO memristor is presented in Fig. 3a. It consists of logic inputs (\({TB}\)), where the input voltage is applied to top (\(T\)) and bottom (\(B\)) electrodes, and the resulting resistance state of the memristor represents the logic output \(Z\). The DC logic concept using a single CuTeHO memristor can be understood by a finite state machine (FSM), as illustrated by Fig. 3b. Condition 1 corresponds to the SET process (V = |Vset|, Icc = 60 nA). In this condition, the Icc of 60 nA slightly increases the RESET voltage of the CuTeHO memristor due to a thicker filament formation, which prevents the partial RESET at −3.5 V (Fig. 3c and Supplementary Fig. 13a). When the initial resistance state \({Z}^{{\prime} }\) is ‘0’, only the logic inputs \({TB}\) of ‘10’ can switch the device into LRS (‘1’). This operation corresponds to \({T}{{{{{\rm{NIMP}}}}}}B\), equivalent to \(T{{{{{\rm{AND}}}}}}({{{{{\rm{NOT}}}}}}B)\). However, \({Z}^{{\prime} }\) does not change at any inputs when it is ‘1’. Both cases should be considered together in condition 1. Furthermore, condition 2 corresponds to the RESET process with a different voltage and Icc (V = |Vreset|, Ireset <Icc <Iset). When a positive Vreset is applied with an Icc of 50 pA, the CuTeHO memristor cannot fully switch from HRS to LRS, resulting in TS behavior due to weak filament formation at this low Icc (Supplementary Fig. 13b). On the other hand, a negative Vreset enables switching from LRS to HRS regardless of Icc level, due to the self-rectifying characteristics of the CuTeHO memristor. When \({Z}^{{\prime} }\) is ‘1’, only the logic inputs \({TB}\) of ‘01’ can switch the device to HRS (‘0’), thus executing \({B}{{{{{\rm{IMP}}}}}}T\), which is equivalent to \(\left({{{{{\rm{NOT}}}}}}B\right){{{{{\rm{OR}}}}}}T\). To summarize, the two voltage conditions can be defined as follows:
The experimentally measured state transition processes of the FSM are displayed in Fig. 3d. All 16 logic operations are experimentally demonstrated by sequentially applying voltage conditions specific to each logic operation in no more than three steps to substantiate the viability of the suggested concept further (Fig. 3e and Supplementary Figs. 14–19). The logic operations can also be performed in a pulse mode by controlling the pulse amplitude and width. The pulse operation of the CuTeHO memristor is provided in Supplementary Fig. 20. Compared to other related studies that conducted logic operations with a single device, the CuTeHO memristor offers advantages of fabrication simplicity and operational efficiency over the one diode-one resistor (1D1R) approach39 or the trilayer-oxide-based unipolar resistive switching (URS) device approach40. Using the suggested DC logic method, a 1-bit binary full adder/subtractor was constructed using a 2 × 2 crossbar in 8 steps (Supplementary Notes 3 and 4, Supplementary Figs. 21–23). Utilizing the structural and functional characteristics of the crossbar, this work shows that the number of required devices and operation steps are lower than previously reported memristor-based adders (Supplementary Table 2). Especially, the suggested method has a significantly lower energy cost than alternatives. Furthermore, the adder and subtractor can be combined in a single 3 × 2 memristive crossbar, allowing concurrent mapping of outputs for both operations (Supplementary Note 4.3, Supplementary Figs. 24–27 and Supplementary Table 3). These proposed DC logic-based transistorless schemes substantially reduce the area and computational costs of in-memory logic operation while enabling enhanced functional density in a single component.
Encryption and decryption of data
Among the 16 logic operations demonstrated in the previous section, an XOR logic operation can effectively encrypt and decrypt data using cryptographic keys41. For instance, the XOR logic operation of the ‘A’-shaped data and cryptographic key converts the original data into encrypted data, which can then be decrypted by another XOR logic operation with the same key (Fig. 4a). Since the CuTeHO device can generate keys using its SET voltage variation, it is possible to encrypt and decrypt data by combining the key generation with 3-step XOR DC logic, both using the CuTeHO devices.
Here, 20 binary letter-shaped data represented by a 4 × 4 pixel grid were flattened into 1 × 16 sized vectors and mapped onto a 20 × 16 memristor array (Fig. 4b). In this case, the applied SET voltage should be higher than 3.5 V to mitigate the impact of SET voltage variation (since the target is an information storage and not PUF generation). The self-rectifying characteristic of the CuTeHO memristor effectively suppresses leakage current when data is mapped. In another 20 × 16 sub-array, which matches the size of the memory array, a set of PUF keys was generated by setting the voltage of each memristor to 2.85 V. These keys can be reconfigured or concealed as needed. Overall, a 20 × 32 hybrid memory-PUF array was created, demonstrating the dual nature of storing deterministic quantities and generating unpredictable quantities.
A 1 × 16 shaped data and a 1 × 16 PUF key were used as inputs to the word lines (WLs) and bit lines (BL) of a 16 × 16 logic operation array to encrypt the selected shape data using the XOR DC logic scheme (Supplementary Fig. 28a). When the voltages are applied to WLs and BLs in the 16 × 16 encryption array, only the cells at the main diagonal experience the appropriate voltage for the XOR operation of the encryption. Therefore, the result of the logic operation is mapped to the main diagonal. After the encryption, the information on the main diagonal was read, replacing the original letter-shaped data. The PUF array can also be concealed via partial RESET, ensuring that the original data and cryptographic key remain hidden until the decryption process (demonstrated in Supplementary Fig. 28b). The overall results of the encryption using 20 different 16-bit letter-shaped data (‘A’ to ‘T’) with 16-bit cryptographic keys (PUF-1 to PUF-20) are displayed in Supplementary Fig. 29. The data and cryptographic keys can be mapped through a microcontroller-based circuit, and the encryption can also be controlled using the same circuit (Supplementary Fig. 30 and Supplementary Note 5). Furthermore, it was examined if mathematical operations can be performed on encrypted data in noise-prone analog computers, without losing the original information (Supplementary Notes 6 and 7). The experiment using analog memristor chips to perform AI inference on encrypted data confirmed that there was minimal degradation of the overall quality of the inference compared to inference performed on unencrypted data. Thus, the encryption-decryption scheme is robust to noise and minor variabilities typical of analog computers.
Table 2 compares the performance between the memristor-based approach in this work (using projected calculations) and a CMOS-based computer. The energy consumption and latency are lower in the memristor-based approach, especially the energy needed to perform encryption/decryption using XOR logic (details provided in Supplementary Note 8, Supplementary Figs. 35 and 36). These comparisons are based on published projections, which may take a few more years to reach volume production, by when CMOS computers would also be improved. These comparisons indicate the competitiveness of memristor-based technologies, while the quantitative estimates may evolve with technological progress.
Discussion
CMOS-based logic and computing have been highly reliable for decades and have employed arbitrary reconfigurability as the backbone of general-purpose computing. However, performance of digital computers is generally saturating42, leading to increased costs for slower improvements in performance. As such, there is a clearly identified need for post-CMOS technologies2,39, which generally promise better performance (for instance, in terms of energy consumption) compared to CMOS approaches. However, a key issue with post-CMOS technologies is their application specificity due to various fundamental reasons, such as their frequent reliance on hard-coded physical processes to perform computing. Such specificity implies the need for discrete accelerating units for different tasks (such as logic, matrix operations, etc.). Therefore, the next wave of research into post-CMOS technologies will need to address functional reconfigurability and application flexibility.
The CuTeHO memristors offer the first steps towards functionally reconfigurable post-CMOS components. These memristors embody fundamentally tunable ion-migration-based kinetics, which are reflected as various device-level functions with opposing requirements. The CuTeHO-based PUF-generating memristors exhibit various such functional features: reconfigurability, concealability, nonvolatility, logic capability, and scalability via passive crossbar arrays. Thus, the CuTeHO memristors not only offer improved performance, but promise functional generality as well. Such application flexibility, especially of post-CMOS devices that can express universal Boolean logic, is a well-recognized need40.
While the performance estimates are based on experimentally calibrated projections and not direct experimental comparisons to entrenched CMOS technologies at full chip scales, these promising projections (among other memristor-based technologies) will aid the down-selection of post-CMOS technologies that will be moved to large-scale production. Many similar comparisons to other memristor technologies will serve this purpose6,28,43. Such large-scale production of post-CMOS technologies is anticipated to be realized within the next decade2,39.
In summary, this work experimentally demonstrated a compact and multifunctional hardware system in a CuTeHO memristive crossbar architecture that can be used for probabilistic (key generation, encryption/decryption) and deterministic (logic operation) computations, which often have opposing requirements—the former requiring unpredictability and the latter requiring predictability. The PUF, which serves as a digital fingerprint or key generation, relies on the stochastic switching behavior of the memristor. In addition, nonstateful logic was implemented based on binary states induced by different biases. These functions are used to demonstrate data encryption and decryption. Performance calculations show a clear advantage in terms of both energy and latency in the demonstrated encryption operation. This work paves the path for multifunctional memristive systems with an increased spectrum of applications.
Methods
Device fabrication
The 4 × 4 memristive crossbar array consisting of CuxTe1-x/HfO2/Pt devices was fabricated with the following procedure (Supplementary Fig. 8). A 10-nm-thick Ti adhesion layer and a 50-nm-thick Pt bottom electrode (BE) were deposited on a SiO2/Si substrate using an electron beam evaporator (SRN-200, SORONA), followed by a lift-off process. A 10-nm-thick HfO2 switching layer was deposited via thermal atomic layer deposition using Hf[N(CH3)(C2H5)]4 and O3 as Hf precursor and oxygen source, respectively. A 50-nm-thick Cu0.3Te0.7 top electrode (TE) was DC sputtered by co-sputtering from Cu and Te targets (07SN014, SNTEK) with a power of 20 W and 55 W, respectively. The power and time of co-sputtering can be adjusted to achieve the desired x value for other CuxTe1-x devices (Supplementary Table 4). Finally, a 40-nm-thick Pt passivation layer was deposited using the electron beam evaporator, followed by the lift-off process. An additional dry-etching system (PlasmaPro System 100 Cobra, Oxford) was used to make the BE contact for the crossbar fabrication.
Electrical measurements
The DC current–voltage (I–V) characteristics of the devices were measured using the semiconductor parameter analyzer (SPA, HP4155B). During the measurement, the TE was biased, and the BE was grounded. The AC pulse measurement was performed using the SPA, pulse generator (PG, Agilent 81110A), and oscilloscope (OSC, Tektronix TDS 684C). An interface built on LabVIEW was used for all electrical measurements.
Data availability
All the relevant data are available from the corresponding authors upon request.
References
Wang, Z. et al. Resistive switching materials for information processing. Nat. Rev. Mater. 5, 173–195 (2020).
Conklin, A. A. & Kumar, S. Solving the big computing problems in the twenty-first century. Nat. Electron. 6, 464–466 (2023).
Woo, K. S. et al. A true random number generator using threshold-switching-based memristors in an efficient circuit design. Adv. Electron. Mater. 5, 1800543 (2019).
Woo, K. S. et al. A combination of a volatile-memristor-based true random-number generator and a nonlinear-feedback shift register for high-speed encryption. Adv. Electron. Mater. 6, 1901117 (2020).
Gao, Y., Al-Sarawi, S. F. & Abbott, D. Physical unclonable functions. Nat. Electron. 3, 81–91 (2020).
Kim, G. et al. Self-clocking fast and variation tolerant true random number generator based on a stochastic mott memristor. Nat Commun 12, 2906 (2021).
Borghetti, J. et al. Memristive switches enable stateful logic operations via material implication. Nature 464, 873–876 (2010).
Jeong, D. S., Kim, K. M., Kim, S., Choi, B. J. & Hwang, C. S. Memristors for energy-efficient new computing paradigms. Adv. Electron. Mater. 2, 1600090 (2016).
Kim, Y. S., Son, M. W. & Kim, K. M. Memristive stateful logic for edge boolean computers. Adv. Intell. Syst. 3, 2000278 (2021).
Goux, L. et al. Influence of the Cu-Te composition and microstructure on the resistive switching of Cu-Te/Al2O3/Si cells. Appl. Phys. Lett. 99, 053502 (2011).
Woo, K. S. et al. A high‐speed true random number generator based on a CuxTe1−x diffusive memristor. Adv. Intell. Syst. 3, 2100062 (2021).
Woo, K. S. et al. Probabilistic computing using Cu0.1Te0.9/HfO2/Pt diffusive memristors. Nat. Commun. 13, 5762 (2022).
Wang, W. et al. Surface diffusion-limited lifetime of silver and copper nanofilaments in resistive switching devices. Nat. Commun. 10, 81 (2019).
Mullins, W. W. Theory of thermal grooving. J. Appl. Phys. 28, 333–339 (1957).
Yoon, J. H. et al. Highly uniform, electroforming-free, and self-rectifying resistive memory in the Pt/Ta2O5/HfO2-x/TiN structure. Adv. Funct. Mater. 24, 5086–5095 (2014).
Yoon, J. H. et al. Pt/Ta2O5/HfO2-x/Ti resistive switching memory competing with multilevel NAND flash. Adv. Mater. 27, 3811–3816 (2015).
Kim, K. M. et al. Low-power, self-rectifying, and forming-free memristor with an asymmetric programing voltage for a high-density crossbar application. Nano Lett 16, 6724–6732 (2016).
Kim, Y. et al. Nociceptive memristor. Adv. Mater. 30, 1704320 (2018).
Kim, K. M. et al. A detailed understanding of the electronic bipolar resistance switching behavior in Pt/TiO2/Pt structure. Nanotechnology 22, 254010 (2011).
Pappu, R., Recht, B., Taylor, J. & Gershenfeld, N. Physical one-way functions. Science 297, 2026–2030 (2002).
Gassend, B., Clarke, D., Van Dijk, M. & Devadas S. Physical random functions. in Proceedings of the 9th ACM Conference on Computer and Communications Security 148–160 (2002).
Guajardo, J., Kumar, S. S., Schrijen, G. J. & Tuyls, P. FPGA intrinsic PUFs and their use for IP protection. in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 4727 LNCS 63–80 (Springer Verlag, 2007).
Edward Suh, G. & Devadas, S. Physical unclonable functions for device authentication and secret key generation. in 2007 44th ACM/IEEE Design Automation Conference (IEEE, 2007).
Herder, C., Yu, M. D., Koushanfar, F. & Devadas, S. Physical unclonable functions and applications: a tutorial. Proc. IEEE 102, 1126–1141 (2014).
Zhang, R. et al. Nanoscale diffusive memristor crossbars as physical unclonable functions. Nanoscale 10, 2721–2726 (2018).
Nili, H. et al. Hardware-intrinsic security primitives enabled by analogue state and nonlinear conductance variations in integrated memristors. Nat. Electron. 1, 197–202 (2018).
Jiang, H. et al. A provable key destruction scheme based on memristive crossbar arrays. Nat. Electron. 1, 548–554 (2018).
John, R. A. et al. Halide perovskite memristors as flexible and reconfigurable physical unclonable functions. Nat. Commun. 12, 3681 (2021).
Dodda, A. et al. Graphene-based physically unclonable functions that are reconfigurable and resilient to machine learning attacks. Nat. Electron. 4, 364–374 (2021).
Gao, B. et al. Concealable physically unclonable function chip with a memristor array. Sci. Adv. 8, eabn7753 (2022).
Huang, P. et al. Reconfigurable nonvolatile logic operations in resistance switching crossbar array for large-scale circuits. Adv. Mat. 28, 9758–9764 (2016).
Adam, G. C., Hoskins, B. D., Prezioso, M. & Strukov, D. B. Optimized stateful material implication logic for three-dimensional data manipulation. Nano Res. 9, 3914–3923 (2016).
Cheng, L. et al. Reprogrammable logic in memristive crossbar for in-memory computing. J. Phys. D Appl. Phys. 50, 505102 (2017).
Kim, K. M. et al. Single-cell stateful logic using a dual-bit memristor. Phys. Status Solidi - Rapid Res. Lett. 13, 1800629 (2019).
Linn, E., Rosezin, R., Tappertzhofen, S., Böttger, U. & Waser, R. Beyond von Neumann - logic operations in passive crossbar arrays alongside memory operations. Nanotechnology 23, 305205 (2012).
Siemon, A. et al. Realization of boolean logic functionality using redox-based memristive devices. Adv. Funct. Mater. 25, 6414–6423 (2015).
Xu, N., Yoon, K. J., Kim, K. M., Fang, L. & Hwang, C. S. Fully functional logic-in-memory operations based on a reconfigurable finite-state machine using a single memristor. Adv. Electron. Mater. 4, 1800189 (2018).
Xu, L. et al. Memristor-based efficient in-memory logic for cryptologic and arithmetic applications. Adv. Mater. Technol. 4, 1900212 (2019).
Theis, T. N. & Wong, H.-S. P. The end of Moore’s law: a new beginning for information technology. Compu. Sci. Eng. 19, 41–50 (2017).
Nikonov, D. E., Li, H. & Young, I. A. Review of simulation methods for design of spin logic. IEEE Nanotechnol. Mag. 17, 37–42 (2023).
Yang, L. et al. Cryptographic key generation and in situ encryption in one-transistor-one-resistor memristors for hardware security. Adv. Electron. Mater. 7, 2001182 (2021).
Witkowski, W. ‘Moore’s Law’s Dead,’ Nvidia CEO Jensen Huang Says in Justifying Gaming-card Price Hike. https://www.marketwatch.com/story/moores-laws-dead-nvidia-ceo-jensen-says-in-justifying-gaming-card-price-hike-11663798618 (2022).
Jeon, K. et al. Purely self-rectifying memristor-based passive crossbar array for artificial neural network accelerators. Nat. Commun. 15, 129 (2024).
Acknowledgements
R. Stanley Williams is gratefully acknowledged for feedback on the experimental results and assistance with the device analysis. A part of this work was supported as part of the Center for Reconfigurable Electronic Materials Inspired by Nonlinear Neuron Dynamics (reMIND), an Energy Frontier Research Center funded by the US Department of Energy (DOE), Office of Science, Basic Energy Sciences. Specifically, reMIND supported efforts from K.S.W. and S.K. on theoretical modeling, simulations, performance calculations, electrical measurements (K.S.W.), and project co-supervision (S.K.). Any subjective views or opinions that might be expressed in the paper do not necessarily represent the views of the US DOE or the United States Government. Sandia National Laboratories is operated for the US DOE’s National Nuclear Security Administration under contract DE-NA0003525. This work was also supported by the National Research Foundation of Korea (Grant No. 2020R1A3B2079882).
Author information
Authors and Affiliations
Contributions
K.S.W. designed the study concept. K.S.W. and J.H. performed electrical measurements. J.H. fabricated the device and refined the concept. S.Y. performed energy and latency calculations. L.T. oversaw the design, assembly and operation of the hardware for AI inference. H.P. contributed to the device analysis. S.K. suggested the encrypted inference concept and performed related experiments and modeling. S.K. and C.S.H. supervised the entire project. K.S.W., J.H., S.K. and C.S.H. wrote the manuscript.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Peer review
Peer review information
Nature Communications thanks Jeetendra Singh, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Woo, K.S., Han, J., Yi, Si. et al. Tunable stochastic memristors for energy-efficient encryption and computing. Nat Commun 15, 3245 (2024). https://doi.org/10.1038/s41467-024-47488-x
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41467-024-47488-x
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.