Temporal correlation detection using computational phase-change memory

Abstract

Conventional computers based on the von Neumann architecture perform computation by repeatedly transferring data between their physically separated processing and memory units. As computation becomes increasingly data centric and the scalability limits in terms of performance and power are being reached, alternative computing paradigms with collocated computation and storage are actively being sought. A fascinating such approach is that of computational memory where the physics of nanoscale memory devices are used to perform certain computational tasks within the memory unit in a non-von Neumann manner. We present an experimental demonstration using one million phase change memory devices organized to perform a high-level computational primitive by exploiting the crystallization dynamics. Its result is imprinted in the conductance states of the memory devices. The results of using such a computational memory for processing real-world data sets show that this co-existence of computation and storage at the nanometer scale could enable ultra-dense, low-power, and massively-parallel computing systems.

Fig. 1

The concept of computational memory. a Schematic of the von Neumann computer architecture, where the memory and computing units are physically separated. A denotes information stored in a memory location. To perform a computational operation, f(A), and to store the result in the same memory location, data is shuttled back and forth between the memory and the processing unit. b An alternative architecture where f(A) is performed in place in the same memory location. c One way to realize computational memory is by relying on the state dynamics of a large collection of memristive devices. Depending on the operation to be performed, a suitable electrical signal is applied to the memory devices. The conductance of the devices evolves in accordance with the electrical input, and the result of the operation can be retrieved by reading the conductance at an appropriate time instance

Fig. 2

Crystallization dynamics. a Schematic of a mushroom-type phase change memory device showing the phase configurations. b Illustration of the crystallization dynamics. When an electrical signal with power P _inp is applied to a PCM device, significant Joule heating occurs. The resulting temperature distribution across the device is determined by the thermal environment, in particular the effective thermal resistance, R _th. The effective thickness of the amorphous region, u _a, evolves in accordance with the temperature at the amorphous–crystalline interface, T _int, and with the temperature dependence of crystal growth, v _g. Experimental estimates of c R _th and d v _g

Fig. 3

Temporal correlation detection. a Schematic of N stochastic binary processes, some correlated and the remainder uncorrelated, arriving at a correlation detector. b One approach to detect the correlated group is to obtain an uncentered covariance matrix. By summing the elements of this matrix along a row or column, we can obtain some kind of numerical weights corresponding to the N processes and can differentiate the correlated from the uncorrelated group based on their magnitudes. c Alternatively, the correlation detection problem can be realized using computational memory. Here each process is assigned to a single phase change memory device. Whenever the process takes the value 1, a SET pulse is applied to the PCM device. The amplitude or the width of the SET pulse is chosen to be proportional to the instantaneous sum of all processes. By monitoring the conductance of the memory devices, we can determine the correlated group

Fig. 4

Experimental platform and characterization results. a Schematic illustration of the experimental platform showing the main components. b The phase change memory array is organized as a matrix of word lines (WL) and bit lines (BL), and the chip also integrates the associated read/write circuitries. c The mean accumulation curve of 10,000 devices showing the map between the device conductance and the number of pulses. The devices achieve a higher conductance value with increasing SET current and also with increasing number of pulses. d The mean and standard deviation associated with the accumulation curve corresponding to the SET current of 100 μA. Also shown are the distributions of conductance values obtained after application of the 10th and 40th SET pulses

Fig. 5

Experimental results. a A million processes are mapped to the pixels of a 1000 × 1000 pixel black-and-white sketch of Alan Turing. The pixels turn on and off in accordance with the instantaneous binary values of the processes. b Evolution of device conductance over time, showing that the devices corresponding to the correlated processes go to a high conductance state. c The distribution of the device conductance shows that the algorithm is able to pick out most of the correlated processes. d Generation of a binary stochastic process based on the rainfall data from 270 weather stations across the USA. e The uncentered covariance matrix reveals several small correlated groups, along with a predominant correlated group. f The map of the device conductance levels after the experiment shows that the devices corresponding to the predominant correlated group have achieved a higher conductance value