Exploiting Memristors for Compressive Sampling of Sensory Signals

Memristors are considered as one promising device for implementing future memory and computing systems. However, design of memristor-based circuits faces a key challenge to deal with variations from nonideal fabrication processes and the resulting performance uncertainties. Fortunately, this randomness can be exploited in many other applications, such as compressive sensing-based data acquisition, which relies on a random sensing matrix in digital or analog formats. Existing compressive sensing systems are usually implemented in digital CMOS circuits, which introduce high hardware complexity and limited sampling speed. In this paper, we studied the process variations in memristor devices and developed an analytical model to evaluate the properties of analog random sensing matrices for compressive sensing applications. Simulations are carried out to demonstrate the statistical distributions of memristor arrays under different filament lengths and identify the optimal switching strategy. The proposed memristor-based compressive sensing circuit also has the advantages of low complexity and high performance. A case study is conducted for compressive image acquisition using the proposed approach.