Memristor-based memory technology, also referred to as resistive RAM (RRAM), is one of the emerging memory technologies potentially to replace conventional semiconductor memories such as SRAM, DRAM, and flash. Existing research on such novel circuits focuses mainly on the integration between CMOS and non-CMOS, fabrication techniques, and reliability improvement. However, research on (manufacturing) test for yield and quality improvement is still in its infancy stage. This paper presents fault analysis and modeling for open defects based on electrical simulation, introduces fault models, and proposes test approaches for RRAMs. The fault analysis reveals that unique faults occur in addition to some conventional memory faults, and the detection of such unique faults cannot be guaranteed with just the application of traditional march tests. The paper also presents a new Design-for-Testability (DfT) concept to facilitate the detection of the unique faults. Two DfT schemes are developed by exploiting the access time duration and supply voltage level of the RRAM cells, and their simulation results show that the fault coverage can be increased with minor circuit modification. As the fault behavior may vary due to process variations, the DfT schemes are extended to be programmable to track the changes and further improve the fault/defect coverage.

Resistive random access memory (RRAM) is one of the universal memory candidates for computer systems. Although RRAM promises many attractive advantages (e.g., huge data storage, smaller form-factor, lower power consumption, non-volatility, etc.), there are many open issues that still need to be solved, especially those related to its quality and reliability. For instance, open defects may cause RRAM cell to enter an undefined state (i.e., somewhere between logic 0 and 1), making it hard to detect during manufacturing test. As a consequence, this may lead to test escapes (quality issue) and field failures (reliability issue). This paper shows - based on defect and circuit simulation - how testing RRAM is different from testing conventional random access memories and how march test cannot guarantee higher defect coverage. The paper then motivates the need of development of special Design-for-Testability (DfT). A concept of a new DfT is then proposed. The concept is further exploited and mapped into two different DfT circuitries: (i) Short Write Time and (ii) Low Write Voltage. Both DfT schemes are implemented and simulated; the simulation results show that defects causing the RRAM cell to enter an undefined state are easily detected.

Hybrid CMOS/memristor memory (hybrid memory)technology is one of the emerging memory technologies potentially to replace conventional non-volatile flash memory. Existing research on such novel circuits focuses mainly on the integration between CMOS and non-CMOS, fabrication techniques and reliability improvement. However, research on defect analysis for yield and quality improvement is still in its infancy stage. This paper presents a framework of defect oriented testing in hybrid memory based on electrical simulation. First, a classification and definition of defects is introduced. Second, a simulation model for defect injection and circuit simulation is proposed. Third, a case study to illustrate how the proposed approach can be used to analyze the defects and translate their electrical faulty behavior into fault models - in order to develop the appropriate tests and design for testability schemes - is provided. The simulation results show that in addition to the occurrence of conventional semiconductor memories faults, new unique faults take place, e.g., faults that cause the cell to hold an undefined state. These new unique faults require new test approaches (e.g., DfT) in order to be able to detect them.

Existing work on fault tolerance in hybrid nanoelectronic memories (hybrid memories) assumes that faults only occur in the memory array and the encoder, not in the decoder. However, as the decoder is structured using scaled CMOS devices, it is also becoming vulnerable to faults. This paper presents a cost-efficient fault-tolerant decoder for hybrid memories that are impacted by a high degree of non-permanent clustered faults. Fault-tolerant capability is achieved by combining partial hardware redundancy scheme and on-line masking scheme based on Müller C-gates. In addition, the cost-efficient implementation of the decoder is realized by modifying the decoding sequence and implementing it based on time redundancy. Experimental results show that the proposed decoder is able to provide better reliability of the overall hybrid memory system, yet requires smaller area as compared to conventional decoder. For example, when assuming the fault ratio between decoder and memory array is 1:10 and at 10% fault rate, the proposed decoder ensures 1% higher reliability of the overall hybrid memory system. Moreover, the proposed decoder realizes 18.4% smaller area overhead for 64-bit word hybrid memory.

Hybrid memories are envisioned as one of the alternatives to existing semiconductor memories. Although offering enormous data storage capacity, low power consumption, and reduced fabrication complexity (at least for the memory cell array), such memories are subject to a high degree of intermittent and transient faults leading to reliability issues. This article examines the use of Conventional Redundant Residue Number System (C-RRNS) error correction code, which has been extensively used in digital signal processing and communication, to detect and correct intermittent and transient cluster faults in hybrid memories. It introduces a modified version of C-RRNS, referred to as 6M-RRNS, to realize the aims at lower area overhead and performance penalty. The experimental results show that 6M-RRNS realizes a competitive error correction capability, provides larger data storage capacity, and offers higher decoding performance as compared to C-RRNS and Reed-Solomon (RS) codes. For instance, for 64-bit hybrid memories at 10% fault rate, 6M-RRNS has 98.95% error correction capability, which is 0.35% better than RS and 0.40% less than C-RRNS. Moreover, when considering 1Tbit memory, 6M-RRNS offers 4.35% more data storage capacity than RS and 11.41% more than C-RRNS. Additionally, it decodes up to 5.25 times faster than C-RRNS.