The high switching speed at sub nanosecond and the excellent endurance of over 10¹² make the ferroelectric memory a competitive candidate for replacing the current flash memory products which suffer from slow speed of ~10^-3 s and limited endurance of ~10⁴ cycles. Currently, there are three basic ferroelectric memory structures namely capacitor-type ferroelectric random-access memory (FeRAM), ferroelectric tunnel junction (FTJ) and ferroelectric field-effect transistor (FeFET).

FeRAM: Ramtron proposed capacitor-based FeRAM in the 1990s. Because of the well endurance and stability, FeRAM products are now competitive for low-volume applications such as smart cards, energy meters, airplane black boxes, microcontrollers and so on.

Issues: The destructive charge read mode limits high-density memory and in-memory computing applications.

FTJ: Experimentally achieved in the 2000s. Its non-destructive conductance read mode and simple structure are appealing for high-density memory and in-memory computing applications.

Issues: Poor endurance (usually <10⁴) due to the ultrathin thickness for direct quantum tunneling hinders its commercialization.

FeFET: Experimentally demonstrated in the 1970s and revived since the discovery of Hf-based ferroelectrics in the 2010s. Its non-destructive conductance read mode is appealing for high-density memory and in-memory computing applications.

Issues: The lack of an epitaxial template results in mesoscopically disordered and polycrystalline ferroelectric on a semiconductor which intrinsically leads to uncontrolled device-to-device variation in nanoscale devices. The variation issue makes the commercialization of FeFET challenging. The three-terminal structure has lower array density compared to two-terminal ferroelectric memory.

Figure 1 Comparison of ferroelectric memory performances.

Solution:

Tian’s team from East China Normal University proposed an innovative ferroelectric fin diode (FFD) in which a ferroelectric capacitor and a fin-like semiconductor channel are combined to share both top and bottom electrodes. The Schottky barrier between the semiconductor channel and bottom electrode renders local field on the vertical semiconductor channel. The bottom electrode act as both gate and drain in a FeFET, resulting in ferroelectric domains-dominated resistive switching.

The FFD memory absorbs merits of both non-destructive conductance read mode as in FTJ and long endurance as in FeRAM while it avoids the non-epitaxial limit which is the source of device-to-device variation in a traditional FeFET.

This innovative design allows ferroelectrics to directly deposit on electrode in an epitaxial way other than on semiconductor, circumventing   the non-uniformity of devices in FeFET. The non-destructive resistive readout of information eliminates the scaling issue in capacitor-based FeRAM where sufficient area is needed for reading charges by domain reversals. The thickness of the ferroelectric defined by channel length does not suffer the direct quantum tunneling limit, avoiding the endurance issue in an FTJ. Furthermore, the self-rectifying property of the FFD permits a passive crossbar array architecture for in-memory computing application without sneak current issue.

Dr. Feng and Prof. Zhu successfully demonstrated FFD memories using different ferroelectric materials (e. g., organic P(VDF-TrFE) polymer and inorganic industrially used Pb(Zr,Ti)O₃ compounds), which emphasizes its universal character. Compared to the vast family of the current nonvolatile memories, this FFD shows high performances such as an endurance of over 10¹⁰ cycles, an ON/OFF ratio of ~10², a feature size of 30 nm, an operating energy of ~ 20 fJ and an operation speed of 100 ns. Benefiting from the simplicity for fabricating this FFD and its self-rectification ratio of ~ 10⁴, a passive crossbar array with 1.6 k units is constructed and used to demonstrate the in-memory computing for a simple pattern classification task. The high device-to-device uniformity is reflected by a small σ/μ value of ~0.023 for positive coercive voltage and ~0.019 for negative coercive voltage using a Gaussian distribution.

Prof. Duan, the lab director, says “this work may arouse great enthusiasm in in ferroelectric and semiconductor communities since it shows great potential for efficient memories and emerging-computing architectures for big data and artificial intelligence applications.”

This work is titled as “A ferroelectric fin diode for robust non-volatile memory” and published in the latest issue of Nature Communications (https://doi.org/10.1038/s41467-024-44759-5).