Atomic Layer-Deposited HfAlOx-Based RRAM with Low Operating Voltage for Computing In-Memory Applications
- 102 Downloads
With Moore’s law closing to its physical limit, traditional von Neumann architecture is facing a challenge. It is expected that the computing in-memory architecture-based resistive random access memory (RRAM) could be a potential candidate to overcome the von Neumann bottleneck problem of traditional computers [Backus, J, Can programming be liberated from the von Neumann style?, 1977]. In this work, HfAlOx-based RRAM which is compatible with CMOS technology was fabricated by an atomic layer deposition (ALD) process. Metal Ag and TaN are selected as top electrodes (TE). Experiments show that the Ag/HfAlOx/Pt device has demonstrated advantages as a memory-computing device because of the low set voltage (0.33~0.6 V) which means low power consumption and good uniformity. Based on a Ag/HfAlOx/Pt structure, IMP logic was implemented at high speed by applying a 100-ns high-frequency low-voltage pulse (0.3 V and 0.6 V). After two steps of IMP implementation, NAND can also be obtained.
KeywordsComputing in-memory RRAM Switching Implemented
Atomic layer deposition
For the boundaries between storage and computing, researchers have proposed a series of research programs: high-bandwidth memory, near-memory computing, and neural compression networks. These methods can reduce the time to access the memory, but they could not solve this problem fundamentally. In order to solve this problem fundamentally, the concept of computing in-memory has gained attention worldwide. It is worth noting that a resistive random access memory (RRAM) device has attracted widespread attention as a competitive candidate for the non-von Neumann computing device because of its capability of in-memory computing [1, 2, 3, 4, 5, 6]. Computing in-memory devices act as both computing and storage units in the same circuit . It was first proposed in 1971 by Chua . Almost 40 years later, RRAM-based logic operation was first proposed in 2010 . Since then, RRAM-based computing in-memory device has been extensively studied and many methods of implementation have been proposed [10, 11, 12, 13, 14]. But as a computing in-memory device, the most crucial feature is stability and low energy consumption. There are still many issues in this area that need to be explored. In this letter, two kinds of RRAM devices were constructed and the electrical properties were tested. In the process of implementing logic operations, stable set and reset voltages and good uniformity between devices are very important indicators.
So far, a wide variety of materials have shown RRAM behaviors, but few of them were compatible with CMOS process. The binary high-k oxides HfAlOx film was deposited using atomic layer deposition (ALD). ALD is well-suited for deposition of oxide films and over layers for various devices and applications  because it is based on surface saturation and precise precursor dosage is not necessary. HfAlOx could be well compatible with the traditional CMOS process and used as the dielectric layer of in-memory computing device. The Ag/HfAlOx/Pt RRAM devices were used to implement stateful logic operations. The IMP logic was regarded as one of four fundamental logic operations (OR, AND, NOT, and IMP) by Whitehead and Russell in 1910 . Moreover, the NAND logic can be obtained by two steps of IMP logic. The NAND logic is known as the universal logic, which means any Boolean logics can be constructed through the NAND logic. This CMOS-compatible, high-speed, and low operation voltage in-memory computing device shows an effective way to solve the traditional von Neumann structure difficulties in the future.
Result and Discussion
Memory and processor are separated in a traditional von Neumann computer architecture . The transfer time of data stored in memory and calculated on the computing unit greatly limits the performance of the computer. It is possible to break the limitation by operating data directly on memory. The research of computing in-memory has the potential to break this limit.
The reason for this phenomenon is that the resistance change mechanism of TaN/HfAlOx/Pt devices is due to avalanching generation and recombination of the oxygen ion and oxygen vacancy dielectric layer. In Ag/HfAlOx/Pt devices, the forming and rupture of conducting filaments, thanks to the redox reactions of metallic Ag, can be driven by a much lower electric field.
s′ ← pIMPs (1).
s′′ ← qIMPs′ (2).
The truth tables showing the equivalence of the sequence of operations to NAND are shown in Fig. 5e.
In summary, two kinds of devices (Ag/HfAlOx/Pt and TaN/HfAlOx/Pt) were fabricated in this study. Both devices show superior switching characteristics. Ag/HfAlOx/Pt device has demonstrated advantages as a computing in-memory device such as CMOS compatibility, good uniformity, low operating voltage, and low power consumption. Logic was implemented through Ag/HfAlOx/Pt RRAM devices. The realization of low operation voltage computing in-memory devices provides an effective way to solve the traditional von Neumann structure difficulties in the future.
This work was supported by the NSFC (61704030 and 61522404), National Science and Technology Major Project (2017ZX02315005), the Program of Shanghai Subject Chief Scientist (18XD1402800), and the Support Plans for the Youth Top-Notch Talents of China.
Availability of Data and Materials
All data are fully available without restriction.
Z-YH prepared the HfAlO-based computing in-memory devices. Then Z-YH and T-YW designed the test methods and assembled the test equipment of in-memory computing. HZ, PZ, and S-JD revised the manuscript. LC, Q-QS, and D-WZ supervised the whole work. All authors critically read and approved the final manuscript.
The authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
- 1.Backus, J. (2007). Can programming be liberated from the von Neumann style?: a functional style and its algebra of programs (p. 1977). ACMGoogle Scholar
- 5.Prodromakis, T. (2013). Two centuries of memristors. In Chaos, CNN, memristors and beyond: a Festschrift for Leon Chua with DVD-ROM, composed by Eleonora Bilotta (pp. 508–517)Google Scholar
- 16.Becker, O. (1962). AN Whitehead and B. Russell, Principia MathematicaGoogle Scholar
- 17.Hennessy JL, Patterson DA (2011) Computer architecture: a quantitative approach. ElsevierGoogle Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.