Fig. 4.
Demo of a single SnOx/SnSe/SnOx device as one core of Markov chain. a The schematic showing the definition of three Markov states and these behaviors are quite related to the initial resistance state. There are five possible filament states insides the SnSe-RRAM: (i) 010, (ii) 121, (iii) 020, (iv) 120, and (v) 021 (numbers represents the conducting filament numbers in upper or lower layers. The first, second, and third numbers represents initial, after set and after reset states, respectively). In order to reduce the complexity of peripheral designs, the five filament states have been classified into three states of Markov chain. b The typical resistive switching behavior for state I: one set and one reset (010), state I: one set and one reset (121), state II: two sets and two resets (020), state III: one set and two resets (120), and state III: two sets and one reset (021). c The relation between initial resistance and the states in 100 cycles. d The probability for each state in 270 testing steps. e The error rate between each state showing continuous decrease after more testing steps. f The Markov chain after validation. The circular with certain stage in it represents the current or next stage. The arrow line between two state represents the state transfer direction and the number near it is the state transition probability between these two states (pji), which is the conditioning probability of next state equal to j when the current state is i [P(Xn+1 = j|Xn = i)]. g Demonstrating the usage of a Markov chain as fixed-probability random number generator. h Validation of the random number generator with error rate lower than 5%. i The four sawtooth waves as one cycle input signal to measure the device with faster speed and lower power. j The generated numbers in ~1.1 × 105 cycles operation of SnSe-RRAM. k The error rate vs. cycle number