Crossbar-Based Memristive Logic-in-Memory Architecture
read more
Citations
Memristive logic: A framework for evaluation and comparison
Memristive Devices for New Computing Paradigms
Memristive self-learning logic circuit with application to encoder and decoder
A Memristive Multiplier Using Semi-Serial IMPLY-Based Adder
Comprehensive Model of Electron Conduction in Oxide-Based Memristive Devices
References
Memristive devices for computing
Metal–Oxide RRAM
Training and operation of an integrated neuromorphic network based on metal-oxide memristors
‘Memristive’ switches enable ‘stateful’ logic operations via material implication
Complementary resistive switches for passive nanocrossbar memories
Related Papers (5)
Frequently Asked Questions (12)
Q2. What is the reason why a different SL geometry would not work?
Given the need to be able to perform as many parallel computations as possible, using a different SL geometry would not work due to current leakage/sneak-paths [32], [34], which contribute to incorrect computations and/or increases in power consumption.
Q3. How many BLs should be used in the logic array?
it is worth noting that, in order to perform two-input logic computations, two BLs from the memory array should correspond to one LIL from the computing array, i.e., there should be 2x as many BLs as LILs.
Q4. What is the way to implement a single-cell repetitive material-stacking?
single-cell repetitive material-stacking structures could enable the implementation of two series ReRAM devices (with the same or opposite polarity) in a single cross-point [29], [30], i.e., 1T2R cells.
Q5. What is the reason why the SLs are not able to perform parallel logic computations?
current from LIL 2 and LIL 4 also flows through the cross-points indicated with dashed squares, which share thesame gated SL, thereby increasing the likelihood of incorrect logic computations.
Q6. Why did the summing amplifiers have the opposite polarity?
Due to the negative sign of the summing amplifier output, the devices in the computing bank had the opposite polarity of that shown in Fig. 2, without loss of generality.
Q7. What are the values of the parameters of the model?
Values of the parameters of the model were set as: {a, b, c, m, f0, L0} = {15000, 0, 0.1, 82, 310}, with ROFF = 200KΩ, and RON = 2KΩ.
Q8. What is the system floor plan shown in Fig. 3?
The system floor plan shown in Fig. 3 includes a 5×6 1T1R memory array at the bottom and a 5×10 1T1R computing crossbar at the top.
Q9. What is the difference in the output voltage amplitude?
The aforementioned difference in the output voltage amplitude can be attributed to the final composite series memristance in each gate, which is higher for the XOR gate; thus, a smaller voltage drops on the RPD2 resistor.
Q10. What are the WLs that were never used?
when driving the rest of the WLs (WL2-4), the authors simultaneously drove iTSL3-5 and oTSL3-5, whereas iTSL/oTSL1 and 6 were never used.
Q11. What was the logic sum of the two products?
when driving WL2, the authors simultaneously drove iTSL2, iTSL3, and oTSL2, etc. Each LIL corresponded to one logic product (logic AND), and the logic sum of the two products was computed in the crosspoint cell of LOL1.
Q12. What is the difference between the MRL and other parallel processing logic design concepts?
Compared to other parallel processing logic design concepts, such as the MRL proposed by Kvatinsky et al. [18], it enables the execution of a wider variety of logic operations based on devices with sharp transitions (filamentary or thresholdtype) instead of linear (or homogeneous) switching devices, which generally respond more slowly to the applied input signals [31].