Chalcogenide Glassy Semiconductors – Could They Replace Silicon in Memory Devices?

K. D. Tsendin

Ioffe Physical-Technical Institute of Russian Academy of Sciences St. Petersburg, Russia

1.   Introduction

The phase-change memory in the chalcogenide glassy semiconductors based on a “crystal-glass” phase transition has been known since the mid-1960’s.¹ The discovery of this phenomenon was not accidental, but was based on an understanding of the semiconducting properties of chalcogenide glassy semiconductors (CGS), which were discovered at loffe institute in 1955.²

Chalcogenide glassy semiconductors are semiconducting glasses incorporating elements of the VI group of the periodic table, such as sulfur, selenium, and tellurium. The ability of Ge-Te glasses to undergo fast crystalline-amorphous phase transition induced by a laser beam is being used in rewritable optical media, such as read/write CDs and DVDs. Moreover, a crystalline-amorphous phase transition may be produced by Joule heating due to an electric current. Accordingly, switching from a high (amorphous) to a low (crystalline) resistance state can be therefore used as a solid-state electronic memory, the more so as the resistance may be changed by many orders of magnitude.

The first generation of memory cells based on current-induced “crystal-glass” phase transitions was fabricated in the late 1970’s. Unfortunately, these first-generation memory cells were prone to frequent data loss and hence unsuccessful. This failure can be attributed to the lacking understanding of the microscopic physical mechanisms underpinning the transition. Today, major semiconductor companies like Intel, Samsung and others are working on flash-type memory devices based on chalcogenide phase-change memory (PCM) cells. Such modern PCM cells are considered one of the promising directions for future nonvolatile memories.

Modern PCM cell operation is based on phase-change properties of a CGS, typically Ge₂Sb₂Te₅ (GST). Such GST-based PCM cells have demonstrated good cycling endurance of 10¹² cycles, making a technology known from the mid-1960’s relevant to nonvolatile memories of the 21st century. This chapter will examine the recent developments in microelectronics that have led scientists and technologists to consider the chalcogenide PCM cells as a possible alternative to the silicon-based flash memory devices.

First, a brief review of CGS switching and memory effects will be presented. Then, the properties of “on” and “off states will be considered in more detail, specially emphasizing the nonlinearity of current-voltage characteristics in the “off’ state. We will show that information recording based on the crystal-glass phase transition induced by an electric field pulse in CGS has the peculiar property that the memory state arises not from a semiconducting but from a metallic state. This metallic state appears due to switching effects in thin films that have strongly nonlinear current-voltage characteristics.

2.   Switching and memory effects

Generally, memory effects in solid-state materials arise from some switching effect.^1,3 A schematic diagram of the switching responsible for the PCM effect in a CGS film is sketched in Fig. 1. Each point of the high-resistance branch (“off state) in Fig. l(a),(b) corresponds to an applied voltage V ≤ V_T, where V_T is a threshold voltage. If the applied voltage exceeds V_T, the material switches from the “off state to the low-resistance “on” state after some delay t_D. The switching time isw of the transition is very small: t_SW t_D- Furthermore, the delay t_D depends exponentially on the (V – V_T) voltage difference and on the film thickness L. In a large device, the “on” state consists of a narrow high-current-density filament. If the material inside this filament does not crystallize for whatever reason – e. g., CGS material with a weak tendency to crystallize, insufficiently high temperature in the filament, relatively small time t_ON in the “on” state, etc. – then for currents less than some on-state holding current I_H, the device will reversibly return to the “off state. Thus, one has a reversible switching effect, illustrated in Fig. 1(a).

Conversely, if the material inside the high-current filament does crystallize because of an appropriately chosen CGS material, a sufficiently high filament temperature and a long enough t_ON, the device will not return to the “off state, as shown in Fig. 1(b). After some set t_SET, the material in the filament will crystallize and remain crystalline form without any applied voltage, leading to the phase-change memory effect (PCM).

Figure 1. Schematic views of reversible switching (a) and PCM (b) current-voltage characteristics (see text).

Schematic diagrams of first-generation and modern CGS memory cells are presented in Fig. 2, while the current-voltage characteristic are presented in Fig. 3.

Let us consider in details the amorphous to crystalline transition (set process) and crystalline to amorphous transition (reset process). To obtain the set (crystalline) state one has to use current and voltage from the set region of Fig. 3 in order to reach the temperature of crystallization T_C, which is of the same order as the softening temperature T_G. Moreover, the crystallizing heat-pulse duration must be rather long for the crystallization to be completed. For the reset process (amorphization), one has to use current and voltage from the reset region of Fig. 3, with the material heated to the amorphization temperature T_A, which is of the order of melting temperature T_M. The reset pulse must be short with a rapid cooling slope in order to retain the amorphous state.

Figure 2. Schematic structures of the first-generation cell and two types of modern cells. The initial layer of the modern cells may be in the amorphous state (A) or in the crystalline state (B).

Figure 3. Current-voltage characteristic of a PCM type A cell of Fig. 2, fabricated from the popular Ge₂Sb₂Te₅ (GST) material.

3.   Properties of the “off, “on”, and memory states

In the “off state, the CGS are semiconducting glasses: σ = σ₀exp(–E*/kT), with an effective activation energy E* = E – f(V/V₀) that depends on the electric field. In the “on” state, E* is approximately equal to zero, so one has the phase transition from semiconductor to metal conductivity, where σ = σ₀. The crystallization or memory set process changes the conductivity to yet another value, σ₀ → σ_C. In other words, the memory state has a different conductivity from the “on” state. In a strong field of ~10⁵ V/cm, the temperature T_TR of the phase transition from semiconductor to metal conductivity falls in the 450–600 K range for such compounds as Si₁₂Te₄₈As₃₀Ge₁₀ and As₂Se₃-4As₂Te₃. Table 1 summarizes the most important optical and electrical properties of the Ge₂Sb₂Te₅ (GST), which is currently considered the most promising CGS material for memory applications.

4.   Comparison of Si-based and chalcogenide-based memory cells

The new generation PCM cells are variously known as phase-change random access memory (PRAM), Ovonic unified memory (OUM, to emphasize the contribution of S. Ovshinsky), and chalcogenide RAM (C-RAM). Figure 4 shows the modern PCM cell of the B type (see Fig. 2). A comparison of first-generation and modern PCM cells from the viewpoint of endurance is shown in Table 2. For the first-generation cell compounds the T_TR of the semiconductor to metal phase transition (E* = 0) is shown, whereas the exact value for GST is not available.

Table 1. Known and estimated values for Ge₂Sb₂Te₅,⁴ where E_μ, E_n, and E are activation energies of mobility, carrier concentration, and conductivity respectively.

Figure 4. Modern PCM cell of B type.⁵

Latest results on modern PCM cells promise that dimensions down to D = 10 nm and L = 10 nm may be achievable.⁴ If so, this would permit the use of 22 nm design rules, which may be at the limit of Si-based flash memory. The PCM cell design shown in Fig. 4 makes it possible to increase the amorphous part of a cell step by step. Then a multilevel PCM cell may be fabricated. The dependence of the resistance R of such a PCM cell on the set current I_SET is shown in Fig. 5.

Let us compare in more detail the performance of new generation PCM cells with Si-based flash memory. The advantages of PCM are the following:

•   information retention is ensured by heavy atoms, rather than light electrons;

•   simplicity of PCM cell fabrication - a single chalcogenide layer between contacts;

•   endurance: Si-based flash memory - 10⁵, whereas PCM cells - 10^11,12;

•   set time: Si-based flash requires 10^-3–10^-4 s, whereas PCM cells can be set in ~10^-7 s (and reset even faster, at ~5 x 10^-8 s);

Table 2. The composition dependence of semiconductor to metal phase transition temperature T_TR and endurance in cycles. The first four rows are from Ref. 7.

Figure 5. Resistance vs. I_SET for a multilevel PCM cell.⁵

•   scaling: Si-based flash cannot go below the 45 nm design rule due to electron tunneling from the floating electrode, whereas PCM cells down to 10 x 10 nm may be manufacturaba,⁴ thus making PCM a possibility for the 22 nm node, unlike Si-based flash;

•   PCM or PRAM has higher performance both because the memory element can be switched more quickly and also because single bits may be changed to either 1 (crystalline) or 0 (amorphous) without erasing an entire block of cells.

Conversely, PCM memory suffers from some disadvantages:

•   the greatest disadvantage of PCM is the high programming current density ~10⁷ A/cm², whereas for a typical transistor this value is ~10⁵–10⁶ A/cm²;

•   another PCM problem is the contact between the hot phase-change region and the adjacent dielectric and metal regions with different thermal expansion properties;

•   another challenge for PCM memory is the slow long-term resistance and threshold voltage drift (~t^0.1);

•   density: while PCM cells may eventually reach the 22 nm (or smaller) node, for now the maximum demonstrated integration of PCM is ~0.5 Gb, as opposed to over 8 Gb for flash.

5.   Physical model of the set-recess process

In order to further improve PCM memory performance, it is necessary to consider the physical nature of the set-reset processes. Modern PCM cell layout and electrode metal have been chosen so that all cell resistance in the memory state is dominated by the bottom electrode. Then this electrode serves as the “heater” for the reset process. Due to this “heater”, a small volume of melted chalcogenide can be quickly cooled to become a glassy semiconductor (short reset pulse of Fig. 3). We shall not consider here the very complex amorphization process, but rather focus on the set process. In this case the cell resistance is due to the high resistance of the CGS material where Joule heating takes place. In this case, it appears likely that the switching and memory effects are determined by electronic-thermal processes. This means that Joule heating, governed by the nonlinear current-voltage characteristic of CGS, is the key to the set process. The nonlinearity takes place in a strong electrical field, and may be described by decreasing of the effective energy of activation E* = (E–f(V/V₀)).

It is known that the electronic-thermal theory of switching can be described by the following equations:

where Q₁ = SLσF², Q₂ = λS(T–T₀), S is the contact area, F is the electric field, λ is the cooling coefficient, and σ = enμ = σ₀exp(–E*/kT).

For f(V/V₀) = 0, we have simple thermal breakdown, Τ_OFF = T_T = T₀(l+kTo/E) and Τ_ON ~ E/k, that is the temperature Τ_OFF at the end of the “off state is very close to the ambient temperature T₀. But the temperature in the “on” state becomes unrealistically high. The situation changes drastically for the electronic-thermal case. For reasons of simplicity, let us consider the case f(V/V₀) = V/V₀, where V₀ = akT. The experimental results give evidence that two extreme points of the S-shaped current-voltage characteristic (CVC) differ only slightly. In this approximation, the solution of the equations (1) gives the following results: T_T ~ T_ON ~ 2T₀, as long as V_T and V_ON do not differ too much. The exact equality T_ON ⁼ 2T₀ holds only if the S-shaped CVC disappears entirely when V_T = V_ON. A small difference between V_T and V_ON yields T_T < 2T₀ and T_ON ≥ 2T₀, but nevertheless T_ON is on the order of 2T₀ rather than T_ON ~ E/k.

Let us consider the case when the phenomenon that determines f(V/V₀) is the multiphonon tunnel ionization of negative-U defects.⁸ The existence of such defects in CGS materials is well established. Figure 6(a) shows the model band diagram of the material, whereas Figs. 6(b) and (c) illustrate schematically the multiphonon tunnel ionization of negative-U defects for D^- – e → D° and D° – e → D⁺ processes, respectively.

The set equations of the model can be solved,⁹ leading to the following approximate result, where τ₂ is characteristic time and q is charge of electron:

We find that the last two terms, i.e. f(V/V₀), are more complicated than our simple V/V₀ approximation. Still, plugging the carrier concentration (2) in the conductivity σ = enμ and solving equations (1) we obtain a CVC plotted in Fig. 7. We find rather good qualitative agreement between the experimental S-shaped CVC and our theoretical prediction based on the multiphonon tunnel ionization processes of negative-U defects. The 650 °C result obtained for Τ_ON is only ~3T₀, where T₀ = 300 K, rather than the much larger E/k.

Figure 6. (a) Energy band diagram of the negative-U centers model, E₁ and E₂ are the first and second energies of the negative-U center thermo-ionization; (b) the potential seen by an electron escaping from negative-U center in the D^– state (where IEI is the energy of the escaped electron); (c) the potential seen by an electron escaping from negative-U center in the D⁰ state, where line 1 indicates the Coulomb potential, line 2 – the external electric field, and E_PF Is the Poole-Frenkel energy barrier lowering.

Figure 7. A comparison of experimental⁶ (black dots) and theoretical (line) CVCs.

Now, we can formulate from the electronic-thermal theory standpoint the reasons why the new generation PCM cells are successful:

•   nanometer dimensions and the rather small difference between V_T and V_TN suppresses current filamentation;

•   nonlinearity of the current-voltage characteristics decreases effective activation energy of conductivity E* = (E – f(V/V₀)) and hence decreases the on-state temperature T_ΟΝ;¹⁰

•   the initial state is crystalline, so only a small part of the film is made amorphous (reset) and goes back to the crystalline state (set);

•   the Ge₂Sb₂Te₅ compound provides a large difference between the retention time of the amorphous state at 300 K (tens of years) and the fast, –100 ns crystallization time at temperature T_C.⁶

All these improvements – nanoscale dimensions of PCM cell, initial crystalline state, and the superior properties of the GST compound – have led to an enormous increase in PCM cell endurance, as shown in Table 2.

5.   Conclusions

Today, investment in PCM research has increased sharply. There is an realistic probability that these efforts will succeed in producing manufacturable PCM-based version of flash memory using the CGS compound Ge₂Sb₂Te₅.^11,12 Unlike the single element Si, the CGS family consists of an enormous number of compounds. Given a sufficiently accurate theory, one can imagine searching the CGS space for materials with relevant parameters – melting and glassy temperatures, energy activation of crystallization, etc. – that will perform even better than Ge₂Sb₂Te5. To summarize, as the title of this chapter notes, it is indeed possible that CGS phase-change memories will replace silicon for nonvolatile applications.

References

1.   S. R. Ovshinsky, “Reversible electrical switching phenomena in disordered structures,” Phys. Rev. Lett. 21, 1450 (1968).

2.   N. A. Gorunova and B. T. Kolomiets, Zh. Tekhn. Fiz. 25, 984 (1955); N. A. Gorunova and B. T. Kolomiets, Zh. Tekhn. Fiz. 25, 2069 (1955).

3.   B. T. Kolomiets and E. A. Lebedev, Radiotekhnika Eletronika 8, 2037 (1963).

4.   T. Kato and K. Tanaka, “Electronic properties of amorphous and crystalline Ge₂Sb₂Te₅ films,” Japan. J. Appl. Phys. 44, 7340 (2005).

5.   A. I. Popov, Physics and Technology of Disordered Semiconductors [in Russian], Moscow: MEI, 2008, p. 282.

6.   A. L. Lacaita, “Phase change memories: State of the art, challenges and perspectives,” Solid State Electronics 50, 24 (2006).

7.   S. A. Smorgonskaya and V. H. Shpunt, Chapter 7 in: K. D. Tsendin, ed., Electronic Phenomena in Chalcogenide Glassy Semiconductors [in Russian], St. Petersburg: Nauka, 1996.

8.   V. N. Abakumov, V. I. Perel’, and I. N. Yassievich, Nonradiative Recombination in Semiconductors (Modern Problems in Condensed Matter Science series, Vol. 33), Amsterdam: North-Holland, 1991.

9.   K. D. Tsendin, “Electro-thermal theory of the switching and memory effects in chalcogenide glassy semiconductors,” phys. stat. sol. (b) 246, 1831 (2009)

10.   T. Gotoh, “Characteristic at high electric fields in amorphous Ge₂Sb₂Te₅ films,” J. Non-Cryst. Solids 354, 2728 (2008).

11.   A. Popov, “Two generation of phase-change memory devices: difference and common problems,” phys. stat. sol. (b) 246, 1837 (2009).

12.   A. V. Kolobov, P. Fons, and J. Tominaga, “Local structure of amorphous Ge-Sb-Te alloys: Ge umbrella-flip vs. DFT simulations,” phys. stat. sol. (b) 246, 1826(2009).