Abstract:Resistive random access memory (ReRAM) is a promising emerging non-volatile memory (NVM) technology that shows high potential for both data storage and computing. However, its crossbar array architecture leads to the sneak path problem, which may severely degrade the reliability of data stored in the ReRAM cell. Due to the complication of memory physics and unique features of the sneak path induced interference (SPI), it is difficult to derive an accurate channel model for it. The deep learning (DL)-based detection scheme \cite{zhong2020sneakdl} can better mitigate the SPI, at the cost of additional power consumption and read latency. In this letter, we first propose a novel CC scheme which can not only reduce the SPI in the memory array, but also effectively differentiate the memory arrays into two categories of sneak-path-free and sneak-path-affected arrays. For the sneak-path-free arrays, we can use a simple middle-point threshold detector to detect the low and high resistance cells of ReRAM. For the sneak-path-affected arrays, a DL detector is first trained off-line (prior to the data detection of ReRAM). To avoid the additional power consumption and latency introduced by the DL detector, we further propose a DL-based threshold detector, whose detection threshold can be derived based on the outputs of the DL detector. It is then utilized for the online data detection of all the identified sneak-path-affected arrays. Simulation results demonstrate that the above CC and DL aided threshold detection scheme can effectively mitigate the SPI of the ReRAM array and achieve better error rate performance than the prior art detection schemes, without the prior knowledge of the channel.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the sneak - path problem in resistive random - access memory (ReRAM), which seriously affects the reliability of data storage in ReRAM arrays. Specifically:
1. **Sneak - path problem**: ReRAM adopts a cross - bar architecture, which can lead to the sneak - path problem (SPI). That is, when reading a certain memory cell, other unselected cells may generate interference currents, thus affecting the reading accuracy and leading to reading errors.
2. **Limitations of existing methods**:
- Traditional model - based methods are difficult to accurately model SPI because SPI has complex data dependencies and memory physical characteristics.
- Although deep - learning (DL) detection schemes can better mitigate SPI, they will bring additional power consumption and read latency.
3. **Proposed new scheme**: To solve the above problems, this paper proposes a new constrained - coding (CC) and deep - learning - assisted threshold - detection scheme, aiming to effectively reduce the impact of SPI while avoiding excessive increases in power consumption and latency.
### Specific solutions
- **Constrained - coding (CC) scheme**:
- A new CC scheme is proposed, which can not only reduce the occurrence of SPI but also effectively divide the memory array into two categories: arrays without sneak - path and arrays affected by sneak - path.
- For arrays without sneak - path, a simple mid - point threshold detector can be used to distinguish between high - and low - resistance - state cells.
- For arrays affected by sneak - path, first, a deep - learning detector is trained offline, and then a threshold detector is derived based on its output for online data detection.
- **Deep - learning - assisted threshold detection**:
- For arrays affected by sneak - path, a deep - learning - based threshold - detection scheme is proposed. The detection threshold is determined by training the deep - learning model offline, thus avoiding the additional power consumption and latency caused by using the deep - learning detector online.
### Main contributions
- **No prior knowledge required**: The proposed scheme does not need to know the specific model of the ReRAM channel in advance, so it is more versatile and adaptable.
- **Performance improvement**: Simulation results show that this scheme can effectively reduce the impact of SPI under different noise levels, and the bit - error - rate (BER) performance is better than existing detection schemes.
- **Reduction in power consumption and latency**: By introducing the threshold detector, the usage frequency of the deep - learning detector is reduced, thereby significantly reducing power consumption and read latency.
In summary, this paper aims to provide an efficient, low - power - consumption solution to deal with the sneak - path problem in ReRAM by combining constrained - coding and deep - learning techniques.