With the explosive growth of digital data in the era of the Internet of Things (IoT), fast and scalable memory technologies are being researched for data storage and data-driven computation. Among the emerging memories, resistive switching memory (RRAM) raises strong interest due to its high speed, high density as a result of its simple two-terminal structure, and low cost of fabrication. The scaling projection of RRAM, however, requires a detailed understanding of switching mechanisms and there are potential reliability concerns regarding small device sizes. This work provides an overview of the current understanding of bipolar-switching RRAM operation, reliability and scaling. After reviewing the phenomenological and microscopic descriptions of the switching processes, the stability of the low- and high-resistance states will be discussed in terms of conductance fluctuations and evolution in 1D filaments containing only a few atoms. The scaling potential of RRAM will finally be addressed by reviewing the recent breakthroughs in multilevel operation and 3D architecture, making RRAM a strong competitor among future high-density memory solutions.

https://iopscience.iop.org/article/10.1088/0268-1242/31/6/063002/pdf

Resistive switching memories based on metal oxides: mechanisms, reliability and scaling

Introduced the concept of floating-gate interference in flash memory cells for the first time The floating-gate interference causes V/sub T/ shift of a cell proportional to the V/sub T/ change of the adjacent cells It results from capacitive coupling via parasitic capacitors around the floating gate The coupling ratio defined in the previous works should be modified to include the floating-gate interference In a 012-/spl mu/m design-rule NAND flash cell, the floating-gate interference corresponds to about 02 V shift in multilevel cell operation Furthermore, the adjacent word-line voltages affect the programming speed via parasitic capacitors

Effects of floating-gate interference on NAND flash memory cell operation

A memory cell array has a first and a second storage area. The first storage area has a memory elements selected by an address signal. The second storage area has a memory elements selected by a control signal. A control circuit has a fuse element. When the fuse element has been blown, the control circuit inhibits at least one of writing and erasing from being done on the second storage area.

Nonvolatile Semiconductor Memory Device

As NAND flash memory manufacturers scale down to smaller process technology nodes and store more bits per cell, reliability and endurance of flash memory reduce. Wear-leveling and error correction coding can improve both reliability and endurance, but finding effective algorithms requires a strong understanding of flash memory error patterns. To enable such understanding, we have designed and implemented a framework for fast and accurate characterization of flash memory throughout its lifetime. This paper examines the complex flash errors that occur at 30-40nm flash technologies. We demonstrate distinct error patterns, such as cycle-dependency, location-dependency and value-dependency, for various types of flash operations. We analyze the discovered error patterns and explain why they exist from a circuit and device standpoint. Our hope is that the understanding developed from this characterization serves as a building block for new error tolerance algorithms for flash memory.

/pdf/error-patterns-in-mlc-nand-flash-memory-measurement-598fgyt3mx.pdf

Error patterns in MLC NAND flash memory: measurement, characterization, and analysis

NAND flash memories have bit errors that are corrected by error-correction codes (ECC). We present raw error data from multi-level-cell devices from four manufacturers, identify the root-cause mechanisms, and estimate the resulting uncorrectable bit error rates (UBER). Write, retention, and read-disturb errors all contribute. Accurately estimating the UBER requires care in characterization to include all write errors, which are highly erratic, and guardbanding for variation in raw bit error rate. NAND UBER values can be much better than 10-15, but UBER is a strong function of program/erase cycling and subsequent retention time, so UBER specifications must be coupled with maximum specifications for these quantities.

Bit error rate in NAND Flash memories

While the performance of flash memory exceeds hard disk drives in almost every category, the cost of flash memory must come down in order to gain wider acceptance in mass storage applications. This paper describes a 3.3 V-only 32 Mb NAND flash memory that achieves not only high performance but also low cost with a 94.9 mm/sup 2/ die size, improved yields, and a simple process with 0.5 /spl mu/m CMOS technology. Die size is reduced by eliminating high voltage operation on the bitlines through a self boosted program inhibit voltage generation scheme. Incremental-step-pulse programming results in a 2.3 MB/s program data rate as well as improved process variation tolerance. Interleaved data paths and a boosted wordline results in a 25 ns burst cycle time and a 24 MB/s read data rate. Maximum operating current is less than 8 mA.

A 3.3 V 32 Mb NAND flash memory with incremental step pulse programming scheme

Emerging application areas of mass storage flash memories require low cost, high density flash memories with enhanced device performance. This paper describes a 64 Mb NAND flash memory having improved read and program performances. A 40 MB/s read throughput is achieved by improving the page sensing time and employing the full-chip burst read capability. A 2-/spl mu/s random access time is obtained by using a precharged capacitive decoupling sensing scheme with a staggered row decoder scheme. The full-chip burst read capability is realized by introducing a new array architecture. A narrow incremental step pulse programming scheme achieves a 5 MB/s program throughput corresponding to 180 ns/Byte effective program speed. The chip has been fabricated using a 0.4-/spl mu/m single-metal CMOS process resulting in a die size of 120 mm/sup 2/ and an effective cell size of 1.1 /spl mu/m/sup 2/.

A 120-mm/sup 2/ 64-Mb NAND flash memory achieving 180 ns/Byte effective program speed

A 120mm^2 64Mb NAND Flash Memory Achieving 180ns/byte Effective Program Speed

Rapidly increasing solid-state mass-storage application areas are requiring low cost, high density flash memories with higher read and program throughputs. This paper describes a 3.3 V-only 64 Mb NAND flash memory fabricated using a 0.4 /spl mu/m single-metal CMOS technology. The read throughput of 40 MB/s is achieved by improving the random access time and by introducing a full-chip burst read. A typical program throughput of 5 MB/s corresponding to 180 ns/byte is achieved by using a narrow incremental step pulse programming (NISPP) technique. A staggered row decoder scheme relaxes layout limitations and improves the random access time.

A 120 mm/sup 2/ 64 Mb NAND flash memory achieving 180 ns/byte effective program speed

Recent work on neural-network accelerators has focused on obtaining high performance in order to meet the needs of real-time applications with vastly different performance requirements, including high precision computation, efficiency for various Deep Learning (DL) layer types, and extremely low power to run always-on applications. Applying a single mode or datatype uniformly across these different scenarios would be less efficient than using different operating modes according to different operating scenarios. For example, super-resolution typically requires FP16 precision for higher image quality, while NNs for face-detection need only INT4 or INT8 precision. Using higher precision than INT8 for face detection would result in higher power consumption. A highly programmable NPU capable of covering the diverse workloads observed in the real world is therefore desired. In this paper, we present a neural processing unit (NPU) optimized with the following features: i) reconfigurable data prefetching and operational flow for high compute utilization, ii) multi-precision MACs supporting INT4,8,16, and float16, iii) a dynamic operation mode to cover extremely low-power or low-latency requirements. These features provide the flexibility needed by real world applications within the power constraints of various product domains.

Suk-Chon Kwon

Papers

A 3.3 V 32 Mb NAND flash memory with incremental step pulse programming scheme

A 120-mm/sup 2/ 64-Mb NAND flash memory achieving 180 ns/Byte effective program speed

A 120mm^2 64Mb NAND Flash Memory Achieving 180ns/byte Effective Program Speed

A 120 mm/sup 2/ 64 Mb NAND flash memory achieving 180 ns/byte effective program speed

A Multi-Mode 8K-MAC HW-Utilization-Aware Neural Processing Unit with a Unified Multi-Precision Datapath in 4nm Flagship Mobile SoC