Ferroelectric memory (FRAM) offers unique capabilities which make it an attractive non-volatile memory choice for many applications^[1−3]. Since Böscke et al. discovered the ferroelectricity in nanoscale HfO₂ in 2011 for the first time^[4], the scaling down problem that plagues traditional ferroelectricity has been greatly promoted. Meanwhile, ferroelectric HfO₂ (FE-HfO₂) starts a new future for memory generations due to its distinctive features include RAM-like, write speeds, low voltage, low power write operation, high cycling endurance lifetime and architectural flexibility^[5−7]. However, certain reliability issues must be satisfactorily resolved before they can be widely applied in commercial memories. Imprint problem, as one of the reliability issues of FE-HfO₂, becomes a critical concern for ferroelectric applications such as ferroelectric memory (FeRAM), ferroelectric field-effect transistors (Fe-FETs), and ferroelectric diode^[8−10]. Take FeRAM as an example, the presence of imprint effect will cause the misreading problem in FeRAM 1T-1C array, as illustrated in Fig. 1, retarding its further development. As the coercive field (E_c) of hafnium-based ferroelectricity is usually quite large (1−2 MV/cm) compared with traditional ferroelectric materials (E_c about 1−10 kV/cm), E_c shift caused by imprint in FE-HfO₂ is more serious problem and poses a huge challenge: On the one hand, although the relative rate of E_c change is not significant, the absolute amount of E_c change may also be significant. On the other hand, the shifted E_c is close to breakdown electric field, which leads to very limited tolerance and bad endurance. Figs. 2(a) and 2(b) show the sense current for bit “1” during read operation at 25 and 100 ℃ and the corresponding sensing charge Q, where Q is the integration of current peaks. At the same integration time, Q value decreases at high temperature. As a result, the sense margin, which is defined as the separation between the highest bit "0" and lowest bit "1", can then decrease and may cause read fails at high temperature as shown in Fig. 2(c). The misreading problem caused by imprint seriously affects the practical applications of FeRAM. Therefore, it is necessary to analyze and solve it with in-depth understanding of the mechanism. There have been some reports on the imprint mechanism of hafnium-based ferroelectric^{[10, 11]}. However, how the imprint of hafnium-based ferroelectricity affects the reliability of the FRAM technology, is still missing.

In this work, we systematically investigated the influence of imprint effect on the reliability of Hf_0.5Zr_0.5O₂ (HZO) ferroelectric capacitors, and found that imprint not only plays an important role in retention characteristic of the HZO device, but also appears in at different times in the ferroelectric cycle. In addition, it can be easily observed from statistical data that the imprint effect can lead to serious E_c shift and remanent polarization (P_r) degradation of ferroelectric units, greatly reducing the reliability of ferroelectric memories. Moreover, by using the invented solution, the misreading problem caused by imprint in circuity could be solved.

FE-HfO₂ capacitor were grown on Si (100) wafer. Fig. 3(a) shows the TEM image of the TiN/HZO/TiN capacitor. The fabrication process of FE devices is as follow (Fig. 3(b)): Firstly, a 20-nm thick TiN bottom electrode (BE) were deposited via physical vapor deposition (PVD) method. Subsequently, 10 nm HZO film were deposited by atomic layer deposition (ALD) at 280 ℃ with a Hf : Zr ratio of 1 : 1 as followed. TEMAH, TEMAZ, and H₂O were served as Hf-precursor, Zr-precursor, and oxygen source, respectively. Latterly, 30-nm thick patterned TiN were deposited via PVD method as top electrode (TE). 40 μm × 40 μm dimension square TEs were utilized for electrical tests. At the end, both HZO capacitors were annealed at 500 ℃ for 60 s in N₂ atmosphere using a rapid thermal annealing system (RTA).

When evaluating FRAM retention reliability, imprint is the most important retention reliability mechanism for FRAM^{[11, 12]}. Ferroelectric memories experience more imprint when exposed to higher temperatures and longer periods of time. However, the limit of imprint in FE-HfO₂ has not evaluated yet. To this end, we selected a 10 nm HZO capacitor which was subjected to baking at 200 ℃ for prolonged test. As shown in Fig. 4, for 10 nm HZO device, E_c gradually shifts towards the positive direction as the baking time was prolonged, and exponential growth of ∆E_c was observed within 3000 min due to the high baking temperature. However, limit imprint was discovered when the baking time was as long as 10⁴ min. The observed phenomenon can be understood by the imprint explanation in our previous work^[11]. It is shown that oxygen vacancies near the interface between the dielectric and electrode are at the heart of imprint in HZO. Since oxygen vacancies were unavoidably generated during the growth processes e.g., element diffusion during thermal annealing, physical bombardment during sputtering top electrode, thus, imprint is inevitably appearing. Nevertheless, the amount of oxygen vacancies in HZO film is limited and will not increase during baking, even though the baking time was increasing to 10⁴ min. It should be noted that although E_c shift by imprint has a limit, it was obtained by baking at a high temperature of 200 ℃ for over 10 000 min, indicating that imprint has always existed in ferroelectric memories and is an inevitable phenomenon.

To further investigate the impact of imprint on the read and write operations for FeRAM, electrical test was conducted. Fig. 5 represents typical read and write operation of 1T1C FeRAM storage units. Considering the capacitor polarization state will not change when logic "0" is read, only the impact of imprint on logic "1" is considered here. Usually, the D-value in polarization between "0" and "1" of FeRAM is defined as the memory window. Hence, considering the reliability issue caused by imprint, we further investigated the degradation of the FeRAM memory in high temperature. In order to investigate the influence of the imprint on ferroelectric devices with different operating voltages, we simulated the reading and writing of FeRAM using ferroelectric capacitors and conducted retention tests. As shown in Fig. 5 (b), each sample was subjected to 1000 cycles to wake up, before sent to the baking test. After that, multiple amplitude pulses (V_read and V’_read) were conducted to HZO devices before and after baking in order to obtain the switching polarization (P_sw and P’_sw) logic "1" state. As the baking time increases, which is shown in Figs. 5(c) and 5(d), the subsequent read "1" operation shows a decrease in remanent polarization, and a significant degradation of the memory window is observed. Therefore, for memory application with low voltage operations or multilevel storage, imprint is a significant threat. Only by increasing the read voltage can the impact of imprint be reduced.

Besides imprint effect, wake-up effect and fatigue, which occur in different cycles period, are also critical reliability issues in ferroelectric capacitors^[13−16]. Usually, as shown in Fig. 6(a), with the number of cycles increases, HZO FE capacitors exhibit three states: 1) Waking-up state—the polarization of HZO devices increases with cycling operation due to the redistribution of oxygen vacancies or the transformation of tetragonal phases into orthogonal phases^{[13, 14]}; 2) Waked-up state—The polarization of FE devices remains relatively stable as the number of cycles increases; 3) Fatigue state—Stress gradually accumulates, generating oxygen vacancies at grain boundaries, leading to a decrease in polarization and an increase in leakage^{[15, 16]}. Typically, imprint measurements were conducted by using waked-up FE devices^{[17, 18]}. However, the measurement of imprint for HZO capacitors during the waking-up and fatigue states has not been evaluated yet. To the end, three devices with different pre-cycles (10¹, 10³, and 10⁵ cycles) are tested as shown in Fig. 6(b). After pre-set, both samples were baked at 150 ℃ in air atmosphere for a period of time to accelerate the imprint and E_c shift. First order reversal curves (FORCs) were carried out for the devices before and after 10³ min baking at 150 ℃. FORCs test is used to measure the coercive field and internal electric field (E_internal) inside ferroelectric capacitors, which is suitable to detect the change of E_internal caused by imprint^{[19, 20]}. Fig. 7 shows the correlation between the imprint and number of cycles before baking. As Figs. 7(b)−7(d) show the P−V hysteresis loops of the MFM capacitor with 10 nm HZO film before and after high-temperature annealing. It can be observed for all samples that E_c gradually shifts towards the positive direction and the P_r decreases drastically as the baking time was prolonged. This E_c shift and P_r decrease is more pronounced in the cases of fatigue state devices according to the statistics in Fig. 7(a). The corresponding FORCs measurements after 1000 min, 150 ℃ annealing for three cases are illustrated in Figs. 7(e)−7(h). As shown in Fig. 7(e), the switching densities for unbaked devices is concentrated around −0.2 MV/cm (E_Bias), which is related to different deposition condition for TE and BE. After high temperature annealing, internal electric field shifted to negative direction for both samples, and E_Bias is located near −0.7, −0.8 and −0.9 MV/cm for less waked-up samples, and well waked-up samples and fatigued samples, respectively (Figs. 7(f)−7(h)). This observed phenomenon is supporting the wake-up and fatigue explanation^[12−14], since cycling operation inevitably causes changes in the number or state of oxygen vacancies which is the role of imprint in HZO ferroelectric thin films.

To get rid of the E_c shift and polarization degradation caused by imprint effect, a sensing circuit is designed to overcome the misreading problem as shown in Fig. 8(a). This circuit can detect the ambient temperature of the FE capacitor, and adjust the V_write according to the temperature. The array is divided into multiple banks and a temperature sensor circuit is added in each bank to better sense the temperature change, as shown in Fig. 8(b). The temperature sensing circuit is designed based on the temperature characteristics of CMOS parasitic diode, which can generate a voltage proportional to temperature. After power-on, the temperature sensor in each bank is activated to generate the temperature signal such as B1, which is then input into a MUX. Memory control selects the temperature signal voltage of the bank according to the address of the target cell being operated, and was compared with the reference voltage. If the temperature is greater than 75 ℃, the working voltage is 3.5 V, otherwise it is 2 V as demonstrated in Fig. 8(c). This solution is simple, fast and low-cost, and can effectively solve the misreading problem caused by the imprint of FE capacitors.

In this work, the relationship between imprint and reliability of HZO-based FeRAM is systematically studied. Based on the electrical tests on extreme high temperature, imprint exists a limit in HZO capacitors but difficult to achieve. Imprint have a significant influence on the retention characteristics of the ferroelectric memory by reducing the memory window. Furthermore, imprint effect is observed during all cycling period, and is more serious in fatigue state. Based on the results, we have proposed a practical solution which could suppress the reliability degradation induced by imprint significantly.