**1. Introduction**

Memristive devices offer a great promise not only for ultra-high density data storage [1] but also for in-memory computing applications [2]. In-memory computing utilizes the synaptic plasticity of the memristors, where they can mimic the biological synapse or neuron (neuromorphic) [3]. A lot of parameter should be considered to achieve high-performance memristor-based artificial synapses (dynamic range, linearity, asymmetry, level of states, etc.) [4]. Much effort has been conducted to improve these parameter, such bi-layering [5], irradiation [6], doping [7], and deposition engineering [8]. Most of synaptic parameters can also be enhanced by tuning the potentiation and depression pulse schemes [5]. Nevertheless, none of these parameters can be performed by memristors that exhibit digital characteristics. On the other hand, several challenges exist in fabricating analog memristors, and, most often, the fabricated memristors exhibit semior pure digital characteristics. In this chapter, we review some important methods to induce analog characteristics by nanofabrication and electrical engineering that

could provide useful insight for the device development engineers to transform their digital memristors into analog.

Resistance switching in memristor devices, set (HRS-to-LRS) and reset (LRSto-HRS) processes, can occur either in abrupt or gradual resistance change. **Figure 1** depicts the schematic of current–voltage and conductance-pulses curves of the abrupt and gradual resistance changes [2]. The abrupt resistance change is a phenomenon where the current or conductance of the resistance states is suddenly changed at a threshold voltage, as depicted in **Figure 1(a)** and **(b)**. Any memristor devices with this digital characteristic tend to have limited capability to exhibit multiple resistance states (multibit performance). Henceforth, this abrupt behavior is also called digital switching. The only possible way to induce multibit performance is by varying the current compliance level to limit the size of the conduction filaments, and thus it controls the amount of the current that can pass through the cell (**Figure 1(b)**). On the other hand, gradual resistance change is a phenomenon where the device does not require any threshold voltage to change the current or conduction of the states, as depicted in **Figure 1(c)** and **(d)**. In this case, any given voltage or electrical pulse stimulus is able to modulate the current or conduction of the states [9]; thus, the number of the states that the memristor can exhibit is equivalent to the number of voltage or pulse stimulus it can response to (**Figure 1(d)**). Henceforth, this gradual behavior is also called analog switching.

**Figure 1.** *Schematic of (a, b) digital and (c, d) analog switching behaviors. Reprinted from [2].*

*Practical Approach to Induce Analog Switching Behavior in Memristive Devices: Digital… DOI: http://dx.doi.org/10.5772/intechopen.98607*

It is important to note that, based on our knowledge, the memristor, which works under unipolar mode (employing the same voltage polarity to set and reset the device), cannot show analog switching due to the rapid Joule heating process. Hence, the memristor should be designed in such a way where the Joule heating should not play a major role in its switching mechanism. Bipolar mode, however, has several variants of mode, such as conventional, complementary, and diode-like bipolar modes. **Figure 2** shows the conventional and complementary bipolar resistive switching in Pt/ZnO/ TiN memristor that was observed by Khan SA et al. [10]. They suggested that the two modes can be exhibited by varying the bias condition; counter-clockwise and clockwise voltage sweeps exhibit conventional (BRS) and complementary (CRS) bipolar modes, respectively, as depicted in **Figure 1(a)** and **(b)**. The efficacy of the modes on

#### **Figure 2.**

*I-V curves of (a) conventional analog bipolar (BRS) and (b) complementary (CRS) resistance switching modes in Pt/ZnO/TiN memristive device. Synaptic potentiation and depression of (c) BRS and (d) CRS. Epoch training accuracy of BRS and CRS devices. (e) Training accuracy of CRS and BRS modes. Reprinted from [10].*

the synaptic performance was studied **Figure 1(c)** and **(d)**. Although both modes can exhibit synaptic plasticity (potentiation and depression characteristics), BRS performs superior linearity than that of CRS and, thus, better training accuracy (**Figure 2(e)**).

Memristors having a complementary or a diode-like mode offer benefit to the circuit integrations, which they do not need to be stacked with an additional select device in the array configuration [11, 12]. However, the exact switching mechanism of the complementary and diode-like bipolar modes is still not fully understood. Likewise, the synaptic properties of the complementary and diode-like memristor devices are still less investigated. Henceforth, based on these reasons, the present chapter only focuses on the device development on the conventional bipolar memristive devices.
