Optical Verification of Physically Unclonable Function Devices Based on Spin‐Orbit Torque Switching

Physically unclonable functions (PUFs) are used for various applications such as anticounterfeiting, authentication, and secret key generation. They are generally evaluated through electrical measurements, requiring much time and effort due to the many electrical connections required. This study proposes an optical verification of spin‐orbit torque (SOT) devices for PUF design. SOT devices have gained much attention because they resist degradation and thermal changes and can generate secret keys with high reproducibility. A simple optical method based on the magneto‐optical Kerr effect is introduced to verify the authenticity of the proposed SOT PUF device. Furthermore, this study assesses the feasibility of using a W/CoFeB/MgO/Ta structure as a PUF. The 24‐bit device shows a reliability of 97.8 ± 1.03%, making it a promising candidate for use as a spintronics‐based PUFs.


Introduction
A physically unclonable function (PUF) secures data in a device using its unique physical characteristics. These characteristics analogous to a human fingerprint include features such as physical layouts, electrical properties of components, and other irreproducible factors. [1,2] Uncontrollable variations introduced during manufacturing processes give rise to unique characteristics between devices. [3] Therefore, the same input applied to different devices produces unique and unpredictable outputs that differ from one device to another. These outputs should be reproducible; the same input applied to the same device should produce identical outputs every time. The input and output are referred to as the challenge and response, respectively. A particular challenge and its corresponding response created by an uncontrollable complex system are referred to as a challenge-response pair (CRP), and are used for device authentication. [4,5] Traditional PUFs based on the complementary metal-oxidesemiconductor (CMOS) have relied on transistor mismatches due to process variations to generate secret keys. However, these PUFs are not ideal for Internet of Things (IoT) applications as temperature fluctuations can affect the threshold voltages of the transistors, and many transistors are required. [6,7] Recently, spinorbit torque (SOT) devices-based PUFs have gained attention due to their high reproducibility, infinite endurance, and thermal robustness in generating secret keys. Many studies on SOT PUFs based on electrical measurements have been reported. [8][9][10][11] SOT manipulates magnetization direction by spin currents in a nonmagnet/ferromagnet device and can be measured optically. [12,13] In SOT switching, we use the spin current generated in the NM layer to alter the magnetization of an FM layer, which can be measured optically. It is possible to determine whether or not the device is a genuine PUF by measuring the magnetization state of the device at different points in time.
We propose a PUF design that integrates a simple optical method based on the magneto-optical Kerr effect (MOKE) to verify the authenticity of the PUF and reduce the number of electrical connections required. MOKE is widely used to study magnetic materials and occurs when light reflects off a magnetic material. The magnetization of the material affects the polarization state of the incident light. [14][15][16] We use the inevitable natural variations in PUF properties originating from fabrication processes to generate unique secret keys. We evaluated the main properties of our SOT PUF devices and analyzed the effect of the number of bits on reliability. We expect the overall reliability to increase as the number of bits increases.

Thin Film Design and Magnetic Property Evaluation
It is essential to design magnetic heterostructures with energyefficient and semiconductor manufacturing-friendly materials. We use tungsten (W) films for our SOT devices for the following reasons. In its metastable -W phase, the W exhibits the largest charge-to-spin conversion efficiency among the transition metals. [17] It is also being widely used in the current semiconductor industry because it is semiconductor manufacturing friendly. We confirm that the W in our device is in its -W phase. [18,19] Figure 1a,b shows the magnetic hysteresis loops and the values of effective magnetic anisotropy constant K u,eff for various postdeposition heat treatment conditions, respectively (see Figure S1, Supporting Information, for full magnetic hysteresis loops). All samples annealed at 300, 400, and 450°C show clear perpendicular magnetic anisotropy (PMA).

Proposed Optical Method to Assess
The proposed optical method for evaluation (Figure 2) uses a MOKE microscope setup and does not require transistors. The device comprises 24 magnetic islands representing 1 bit each on a single current path. The magnetic islands have a cylindrical shape with a diameter of d = 20 μm and a height of h = 3.9 nm and consist of CoFeB/MgO/Ta layers. We analyzed the magnetization states after applying each current injection to the device. Since our model differs from a conventional circuit and does not require transistors, we are cautious about using the term supply voltage (i.e., V DD and V CC ). Instead, we employ I write (denoted I w hereafter), which represents the current (voltage) applied to the device. The term I w plays a similar role as V DD in conventional integrated circuits.

Assessment of PUF Properties
We exploit the natural variations in the electrical characteristics of the SOT PUF devices due to the uncontrollable variabilities introduced during the manufacturing process. Figure 3a depicts the optical images of representative devices. Each device comprises 24 magnetic islands composed of an FM layer. Each magnetic island represents one bit since the FM layer with PMA exhibits up or down magnetization states, each corresponding to "0" or "1", respectively. Thus, our devices generate an 8 × 3 = 24-bit-long CRP.
For the following evaluations, we classify the charge current I w as a challenge for the following factors. In general, a particular challenge should result in the same response, and the change in the challenge will alter its corresponding response. This fundamental concept of CRP also applies to charge current, as a constant charge current will reproduce the same SOT switching behavior. In contrast, a change in charge current will vary its corresponding response. Thus, we define charge current as a challenge and analyze its corresponding SOT behavior, which varies from device to device mainly due to uncontrollable physical features introduced during fabrication.
We obtain 24-bit long CRPs (secret keys) by conducting set/reset cycles. First, we conduct a reset operation by applying a magnetic field normal to the plane of the device by an electromagnet to saturate the magnetization of the magnetic islands toward the same direction. It will be the initial state of the device. Then, we carry out a set operation by applying I w as a challenge under appropriate environmental conditions. I w will induce SOT with the aid of an external field. We obtain MOKE images throughout the entire set operation and evaluate them to acquire 24-bit long CRPs. Figure 3b,c shows MOKE images and the corresponding 24-bit CRPs. The light (dark) contrast of magnetic islands in all MOKE images indicates down (up) magnetization states. We classify and assess the key properties of the devices based on the previous study. [20] First, "reliability" is a vital aspect of PUFs. It represents the ability of a PUF to reproduce the same responses when subjected   to the same challenge under varying conditions. To evaluate reliability, we used the following equation: where R i is the n-bit response from device i at a normal operating condition. This serves as the reference response of the system. We obtain the same n-bit response at a different operating condition with a value R ' i . A total number of m samples of R ' i are evaluated, with R ' i,t being the t-th sample of R ' i . The reliability is then assessed by: The more the PUFs are vulnerable to varying environmental conditions, the lower the reliability. A highly reliable PUF should be able to reproduce identical responses under fluctuating environmental conditions.
We classify the applied external field (H x ) as the varying environmental condition for the following reasons. First, external magnetic fields are necessary for deterministic SOT switching. [21,22] As mentioned, enough spin current with an external field can reverse magnetization. The SOT itself cannot deterministically reverse the magnetization but can only orient the magnetization to the in-plane direction. Therefore, if the charge current is injected without an external magnetic field, the magnetization direction remains in the metastable in-plane orientation. When the current is turned off, it has a 50% probability of pointing upward or downward. [23][24][25] A reliable PUF should be able to reproduce responses, which means that such randomness of stochastic magnetization reversal should be eliminated; we operate our devices under the application of external magnetic fields. Also, even though external fields act globally on the devices, there is no guarantee that the same amount will apply to all of them. Therefore, we should consider the degree to which the fluctuation in the external magnetic field hinders the reliability of PUF devices. For these reasons, we select the external magnetic field as an environmental condition and evaluate our devices under such fluctuations. reliability of 97.8 ± 1.03%, ensuring that the devices can reproduce the same responses corresponding to its challenges. Note that the ideal reliability of 100% is hardly ever reached, even with the state-of-the-art CMOS PUFs.
The "uniqueness" of a PUF represents the degree to which PUFs are distinguishable from the same type. Ideally, the same PUF design implemented on multiple devices should produce a uniqueness of 50%. We use Hamming distance (HD) between a pair of PUF responses to obtain uniqueness. We calculate uniqueness by the following: where R i and R j represent the n-bit responses of two devices i and j (i ≠ j), respectively, among k devices. Figure 4b demonstrates the uniqueness of SOT PUF devices obtained by evaluating the responses measured at normal operating conditions. The uniqueness tends to approach the ideal value as the number of bits increases. Although the uniqueness of 61.67% for the 24-bit device may seem somewhat deviated from its ideal value, we anticipate from the tendency of the results as the number of bits increases that further developments in testing protocols or PUF models will enhance the overall uniqueness in the future. The distribution of "0" s and "1" s in the response bits determines the uniformity of the devices. For example, a genuine PUF should produce the same number of "0" s and "1" s in its responses, leading to a uniformity of 50%. We define uniformity as follows: where r i,l represents the l-th binary bit of an n-bit response from device i. We use responses taken at normal operating conditions to calculate the mean uniformity of SOT PUF devices. Figure 4c confirms that the devices have an even number of "1" s and "0" s. Figure 5 presents the CCD images of 24-bit SOT PUF devices in their normal operating conditions. Movies S1-S5 (Supporting Information) show real-time imaging of secret key generation of 24-bit SOT PUF devices in their normal operating conditions. The contrast transition from dark to light indicates SOT switching. Orange boxes in Figure 5a indicate the area used for evaluation, and we average the contrast values corresponding to each horizontal position for evaluation. Average contrast values above (under) a specific value are then converted to 1 (0), which indicates different magnetization states. Figure 5b shows the converted results of the orange boxes in Figure 5a. Magnetic islands are not counted as bit "1" for property evaluation if only a tiny portion of the contrast is 1. Partial magnetization switching is inevitable as micrometer-sized magnetic domains are multidomain. Our results show that light and dark magnetic islands are distinguishable, indicating that optical evaluation of PUF is possible.

Reconfigurable SOT PUF for Thermal Fluctuations
Reliable PUF devices should be able to reproduce the same responses under harsh environments. For example, temperature fluctuation is one of the most common varying environmental conditions electronic devices confront. We present a reconfigurable SOT PUF impervious to temperature fluctuation by establishing a relationship between challenge and temperature. Figure 6a shows the correlation between the challenge required to obtain a uniformity of 50% for individual devices and the operating temperature. I w,avg represents the average value of I w required to obtain a uniformity of 50% for individual devices at the indicated temperature. This relationship ensures that the devices possess an average uniformity of 50% at different operating temperatures.
We inject an appropriate challenge to produce a uniformity of 50% at various temperatures by utilizing the relationship and evaluate the reproducibility of the devices. Figure 6b demonstrates the capability of the devices to reproduce the secret keys at various temperatures. We assign a reference response for every temperature and evaluate reliability properties accordingly. As a result, our devices do not suffer from elevated temperatures; they can reproduce the secret keys irrespective of the operating temperature. We also show that the relationship is fully reversible, as the reliability retains its high reproducibility when the devices are cooled back to room temperature. This relationship enables our SOT PUF devices to function with the required properties regardless of the operating temperature. We also confirm that the devices can maintain their properties after undergoing multiple thermal cycles (see Figure S2, Supporting Information).

Conclusion
In this study, we developed and evaluated a SOT-based PUF model to better understand the impact of device fabrication variations on the extrinsic properties of PUFs. We also introduced a more cost-effective and efficient method for fabricating and testing devices using optical evaluation. Our model proved to be a reliable secret key generator with a 24-bit device acquiring its maximum reliability of 97.8 ± 1.03%. Additionally, we examined the effect of the number of bits on reliability. As the number of bits increases, the impact of a single error bit on reliability decreases. To demonstrate this, we fabricated and compared the reliabilities of 4-bit, 8-bit, and 24-bit SOT PUF devices under external field fluctuations. As expected, the reliability of 4-bit devices decreased significantly compared to the 8-bit and 24-bit devices as external field variation increased.
Measurements: The hysteresis loops of the samples were measured using a vibrating sample magnetometer (VSM, Microsense EV9). A PUF assessment was performed with the MOKE microscope setup. The setup can apply current using a source measure unit (SMU, SMU 2400) and external magnetic fields in the x-or z-axis using electromagnets. It allows real-time imaging of magnetic domains using a CCD camera while applying current or magnetic fields. The setup could visualize magnetic domains at the micrometer scale.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.