betaFeedback

WorksPhysics and Astronomy

Filamentary Transport and Thermoelectric Effects in Mushroom Phase Change Memory Cells

Abstract

We performed a 2D finite-element electrothermal computational study of thermoelectric effects and filamentary electronic transport in Ge2_2Sb2_2Te5_5 mushroom phase change memory cells during Reset and Set operations, accounting for spatial activation energy variations in amorphous Ge2_2Sb2_2Te5_5 and phase-change dynamics. Reset operations with current going from the top electrode to the narrow 4 nm bottom electrode require ∼\sim3x less energy and power, and ∼\sim2x lower current to achieve the same Reset resistance, compared to the opposite polarity, due to thermoelectric effects. Filamentary conduction, electrical breakdown, thermal runaway, and local crystallization of amorphous Ge2_2Sb2_2Te5_5 depend on current polarity and thermal boundary conditions, and determine the location, shape, and volume of the programming region, which may be significantly smaller than the semi-cylindrical mushroom region. The programming volume does not scale with contact dimensions larger than 10 nm. Larger contact areas introduce increased device-to-device and cycle-to-cycle variability due to filamentary conduction but are expected to lead to higher reliability and endurance.

Introduction

In the era of artificial intelligence, big data and neuromorphic computing, the demand for high-speed, energy efficient and scalable memory technologies has surged beyond the capabilities of conventional memory [1]. The performance bottlenecks in von Neumann architectures largely stem from the latency gap between the CPU, volatile DRAM, and non-volatile flash memory. Non-volatile resistive memory (RRAM) technologies such as memristors, magnetic RAM (MRAM) and phase change memory (PCM) emerged as possible candidates to bridge this gap. PCM is the most mature among these [2] and it is in high volume production [3], [4].

The programming region of PCM devices is composed of a phase change material, which can be recursively switched between its crystalline (low-resistance) and amorphous (high-resistance) phases by self-heating with short duration voltage pulses [5]. The programming region is amorphized by heating above melting temperature (TmeltT_{\mathrm{melt}}), followed by rapid quenching, and recrystallized by heating above glass-transition temperature (TglassT_{\mathrm{glass}}).

Authors are with the University of Connecticut, CT 06269 USA (e-mail: [email protected]).

PCM can be SET and Reset with ∼1\sim1–100100 ns voltage pulses in the ∼1\sim1 V range, achieving large resistivity contrast (10210^2–104×10^4\times) in a CMOS compatible resistance range (kΩ\mathrm{k\Omega}–MΩ\mathrm{M\Omega}) and >10> 10 years retention at CPU operating temperatures [6], [7]. PCM can be Reset, Set and read using unipolar voltage pulses (without switching the current direction) and is scalable down to ∼5\sim5 nm, unlike memristors and MRAM [8]. Read operation is non-destructive, performed using small voltage pulses (∼0.1\sim0.1 V), and can be very fast (<1< 1 ns). Current, power and energy requirements for Reset and Set operations scale with phase change material and device size. Ge2Sb2Te5\mathrm{Ge_2Sb_2Te_5} (GST) is widely adopted as the phase change material due to its low thermal conductivity, and suitable melting and crystallization kinetics [5], [9]. Germanium-rich GST, with a higher TglassT_{\mathrm{glass}}, is used for high-temperature applications [3]. Cross-point (sandwich-cell) PCM devices integrated with Ovonic Threshold Switch (OTS) access devices have been mass-produced using 20 nm technology as 3D Xpoint [2], [6], [10]. Mushroom cells, constructed with a common top contact and small bottom contacts, have programming regions surrounded by crystalline phase change material and do not require nucleation for crystallization, leading to higher-speed, and offer the flexibility to use side-wall processes to fabricate very narrow bottom contacts (Wheater∼2W_{\mathrm{heater}} \sim2–3 nm) [11]. If a side-wall process is used for contact definition, rows of devices are formed by lithographically defining and etching the common top-contact and phase change layer, defining the width of the phase change material (out of plane dimension of Figure 1), and the programming regions form as semi-cylinders [8].

PCM cell with a MOSFET in series for top and bottom bias polarity

Figure 1. PCM cell with a MOSFET in series for (a) top and (b) bottom bias polarity. Crystal orientation angles for crystalline phase and current density profiles for amorphous phase of PCM are shown. Wheater=4W_{\mathrm{heater}} = 4 nm.

PCM cells experience high temperatures (T∼900T \sim900 K), extreme current densities (J∼100J \sim100 MA/cm2^{2}) and thermal gradients (∼50\sim50 K/nm) which make the contribution of thermoelectric effects very significant. The Set operation relies on electrical breakdown of the amorphous region, giving rise to highly non-linear current-voltage characteristics that are strong functions of temperature [12]. Hence, understanding the electronic conduction within the amorphous region and at its interfaces is critical for device and waveform design.

Trap-assisted tunneling, Poole–Frenkel transport, and field induced delocalized tail states models have been proposed as electronic conduction mechanisms within amorphous phase change materials [13, 14]. The significant band offsets at the amorphous phase change materials’ interfaces are expected to lead to energy barriers for electrons and holes. Local variations in the disordered amorphous phase give rise to significant variations in current density, local self-heating, filamentary transport and thermal runaway [6]. In our recent modeling study, we have incorporated local variations in amorphous (a-)GST properties, going beyond the commonly used effective-media approximation for a-GST, and demonstrated filament formation in a sandwich-cell structure [6]. In this work, we investigate filamentary conduction, impact of thermoelectric effects and the phase change dynamics during Reset and Set operations of GST mushroom cells [12].

Modeling Framework

We use crystalline and metastable a-GST properties to study the cell behavior during high-speed cycling. Device-scale high-speed low-field resistivity versus temperature (ρ\rho–TT) measurements of metastable a-GST show a simple exponential response to temperature [15] with a trend that matches the molten resistivity of GST at TmeltT_{\mathrm{melt}} [16] (Figure 2). Assuming an Arrhenius behavior, we calculate the effective activation energy (EAE_A) as a function of temperature from the fit parameters ρ1=35137\rho_1 = 35137 Ω⋅\Omega\cdot cm and α=0.0202\alpha= 0.0202 K−1^{-1}, extracted from the experimental data [6, 17] (Figure 2):

Temperature dependent metastable resistivity plot

Figure 2. Temperature dependent metastable resistivity ρ\rho (blue spheres) of amorphous GST measured at device level along with crystalline resistivity. Adapted from [17], licensed under CC BY 4.0.

ρ=ρ1e−αT=ρ0eEA(T)kBT(1)\rho= \rho_1 e^{-\alpha T} = \rho_0 e^{\frac{E_A(T)}{k_B T}} \tag*{(1)}

Assuming EA(Tmelt)=32kBTmeltE_A(T_{\mathrm{melt}}) = \frac{3}{2} k_B T_{\mathrm{melt}}, we obtain:

ρ0=ρ1e−32−αTmelt(2)\rho_0 = \rho_1 e^{-\frac{3}{2}-\alpha T_{\mathrm{melt}}} \tag*{(2)}

From Eqs. (1) and (2), EA(T)E_A(T) can be written as:

EA(T)=kBT{32+α(Tmelt−T)}(3)E_A(T) = k_B T \left\{ \frac{3}{2} + \alpha(T_{\mathrm{melt}} - T) \right\} \tag*{(3)}

EA(T)E_A(T) reaches zero at around 930 K (Figure 2), which we describe as the metal transition temperature (TmetalT_{\mathrm{metal}}) of GST. Temperature-dependent (85 to 300 K) high-field (0 to ∼50\sim50 MV/m) I-V sweeps performed on a-GST line-cells show distinct low-field (< 20 MV/m) and high-field (> 20 MV/m) responses [18]. High-field stress significantly accelerates resistance drift, and the devices stabilize within a few minutes. The subsequent I-V sweeps show stable device characteristics (Figure 3) [18]. The low-field characteristics show hyperbolic sine behavior, which can be attributed to trap assisted transport of electrons (Figure 4). The net current can be written as

I-V characteristics of stable amorphous GST line cells

Figure 3. I-V characteristics of stable amorphous GST line cells with W×L×t≈152W \times L \times t \approx152 nm ×\times 710 nm ×\times 20 nm measured at 85, 125, 175, 225, and 300 K. Adapted from [18].

Band diagrams and schematic conduction-band edge

Figure 4. (a) Room temperature band diagram of TiN/c-GST/a-GST/TiN structure under a mild bias of 0.5 V for 20 nm GST length with illustrative charge traps. The de-trapped holes cannot escape the potential well formed, and only electrons can contribute to transport. (b) Schematic conduction-band edge along transport direction. The trapped charges activation energy (EAE_A) is the peak energy barrier at zero applied bias minus the modulation by the applied bias ΔEA(V)\Delta E_A(V) which depends on the ratio of the peak location (dpeakd_{\mathrm{peak}}) and trap separation (dtrapd_{\mathrm{trap}}).

the difference between the forward and reverse currents that describe thermionic emission over an energy barrier modulated by the applied voltage [19]:

I(V,T)=Iforward−Ireverse=I0e−EA−bωVkBTkBT−I0e−EA−(b−1)ωVkBTkBT(4)I(V,T) = I_{\mathrm{forward}} - I_{\mathrm{reverse}} = I_0 e^{-\frac{E_A-b\omega V k_B T}{k_B T}} - I_0 e^{-\frac{E_A-(b-1)\omega V k_B T}{k_B T}} \tag*{(4)}

where I0I_0 is a current scaling factor, b=dpeak/dtrapb = d_{\mathrm{peak}}/d_{\mathrm{trap}} and ω=qdtrap/(kBTL)\omega= qd_{\mathrm{trap}}/(k_B T L) are parameters extracted from fitting Eq. (4) to low-field experimental data in Figure 3. LL is the length of the measured GST line cell, dtrapd_{\mathrm{trap}} is trap-to-trap distance, and dpeakd_{\mathrm{peak}} is the trap to barrier peak distance for forward transmission (Figure 4). The activation energy (EAE_A) is defined as the peak barrier height when no bias is applied. We observe bb to decrease as a function of temperature and approach ∼0.5\sim0.5 for T>300T > 300 K [19], indicating that the barrier peak is approximately at the center position between the two traps. Expressed in terms of current density JJ, the model for T>300T > 300 K simplifies to:

J(V,T)=I(V,T)cross-section area=J0e−EAkBT(eωV2−e−ωV2).(5)J(V,T) = \frac{I(V,T)}{\text{cross-section area}} = J_0 e^{-\frac{E_A}{k_B T}} \left(e^{\frac{\omega V}{2}} - e^{-\frac{\omega V}{2}}\right). \tag*{(5)}

Taylor series expansion of Eq. (5) yields:

J(V,T)=J0e−EAkBT{[1+ωV2+(ωV/2)22+⋯ ]−[1−ωV2+(−ωV/2)22+⋯ ]}(6)J(V,T) = J_0 e^{-\frac{E_A}{k_B T}} \left\{ \left[1+\frac{\omega V}{2}+\frac{(\omega V/2)^2}{2}+\cdots\right] - \left[1-\frac{\omega V}{2}+\frac{(-\omega V/2)^2}{2}+\cdots\right] \right\} \tag*{(6)}

The low-voltage behavior can be modeled by neglecting the third and higher order terms and expressed in terms of electric field, E=V/LE=V/L, as:

J(V,T)=J0e−EAkBTωV=J0e−EAkBTωLE=1ρE.(7)J(V,T)=J_0 e^{-\frac{E_A}{k_B T}}\omega V=J_0 e^{-\frac{E_A}{k_B T}}\omega L E=\frac{1}{\rho}E. \tag*{(7)}

Hence, the electrical resistivity and J0J_0 can be formulated as:

ρ=ρ0eEAkBT=1ωLJ0eEAkBT(8)\rho=\rho_0 e^{\frac{E_A}{k_B T}}=\frac{1}{\omega L J_0}e^{\frac{E_A}{k_B T}} \tag*{(8)}

where J0J_0 is the current density for the no-barrier-for-forward-transmission case, EA/(kBT)−ωV/2=0E_A/(k_B T)-\omega V/2=0, which signifies the end of the transport limited by thermionic emission: using the ρ(T)\rho(T) and EA(T)E_A(T) extracted from the metastable a-GST measurements (Figure 2) and ω\omega extracted from the stable measurements (Figure 3), we calculate J0=188J_0=188 MA/cm2^2, which corresponds to a maximum velocity (J0/qnJ_0/qn) of ∼105\sim10^5 cm/s for metastable a-GST, assuming carrier concentration n=1022n=10^{22} cm−3^{-3} [19]. As a comparison, carrier saturation velocity in amorphous silicon is 1.7×1041.7\times10^4 cm/s [20] and in crystalline silicon is 1×1071\times10^7 cm/s [21].

The electronic transport in crystalline (fcc) GST is linear in nature (J=E/ρJ=E/\rho) with ρ(300 K)=6.9×10−5 Ω⋅m\rho(300\ \mathrm{K})=6.9\times10^{-5}\ \Omega\cdot\mathrm{m} [22] [23] (Figure 2).

We model the thermal transport during device operation as:

dCpdTdt−∇⋅(k∇T)=−∇⋅(J⋅E)−∇⋅(JST)+QH(9)d C_p\frac{dT}{dt}-\nabla\cdot(k\nabla T)=-\nabla\cdot(J\cdot E)-\nabla\cdot(JST)+Q_H \tag*{(9)}

where dd is the mass density, CpC_p is the specific heat, QHQ_H is the latent heat of phase change, SS is the Seebeck coefficient [23] and kk is the thermal conductivity, 0.27 and 0.4 W m−1^{-1} K−1^{-1} at 300 K for amorphous and crystalline GST respectively [24].

SS and QHQ_H are temperature dependent. Modeling of QHQ_H can be found in [25]. The thermal transport model is solved along with electronic transport and phase change physics model [26] in a fully coupled manner to capture the dynamics of device operation. Electric potential, current density, temperature, and nucleation-growth-amorphization of the GST crystal grains are calculated dynamically as the simulation progresses. A rate equation is used to track crystallinity, which can be generally described as:

∂CD∂t=Nucleation+Growth+Amorphization(10)\frac{\partial C_D}{\partial t}=\text{Nucleation}+\text{Growth}+\text{Amorphization} \tag*{(10)}

where the nucleation term randomly generates nuclei depending on the local temperature. The growth and amorphization terms account for the increase and decrease in crystal grain sizes respectively. A single growth velocity versus temperature function, with a negative growth velocity for T>TmeltT>T_{\mathrm{melt}}, is used to model crystallization-amorphization velocity at the grain boundaries [26].

We model variations in a-GST, which may be due to variations in composition, defects and trapped charges, by randomly varying the EA(T)E_A(T) extracted from the experimental studies using a random distribution function across the GST domain (Figure 5). In the studies we present here, we have used 30% standard deviation in EA(T)E_A(T) in a 2D GST domain with 2 nm ×\times 2 nm pixel size that is smoothed using a diffusion function to reduce the numerical complexity in the simulations.

Random and diffused activation-energy maps

Figure 5. Random EAE_A map with 2 nm ×\times 2 nm granularity in a 100 nm ×\times 20 nm GST region (left) and the diffused EAE_A used for simulations which reduces the EAE_A gradients to enable convergence (right). The heater width is 4 nm.

∂EA∂t+∇⋅(−c∇EA)=0(11)\frac{\partial E_A}{\partial t}+\nabla\cdot(-c\nabla E_A)=0 \tag*{(11)}

2D Modeling Results

Mushroom cells are typically fabricated as vertical cells with titanium nitride (TiN) contacts. Reset is performed with short and high amplitude pulses that melt GST in the programming region. Rapid termination of the pulse, quenching, results in amorphization. Set is typically performed using longer duration pulses, maintaining T>TglassT>T_{\mathrm{glass}} in the programming volume, to enable crystallization. Read is performed using small (∼0.1\sim0.1 V) short duration pulses that do not disturb the state of the cell. The resistance contrast after Reset and after Set operations, Rreset/RsetR_{\mathrm{reset}}/R_{\mathrm{set}}, is referred to here as the memory window (Figure 8).

Cell resistances after Reset and Set for five cycles

Figure 8. (a) Cell resistances after Reset (RresetR_{\mathrm{reset}}) and Set (RsetR_{\mathrm{set}}) for five cycles. (b) A small increase in IresetI_{\mathrm{reset}} and ISetI_{\mathrm{Set}} results in a large increase in Rreset/RsetR_{\mathrm{reset}}/R_{\mathrm{set}}.

As 3D simulations are computationally expensive, we performed 2D geometry simulations for two cross-sections: typical 2D mushroom-cell simulation (xx–yy plane) with 10 nm out-of-plane depth, and out of mushroom-plane (yy–zz plane) simulations with 4 nm out-of-plane depth (Figure 6). We choose the applied voltages to keep the access transistor in linear mode during Read and in saturation mode during programming (Set, Reset) operations (Figure 7).

3D extruded view of the mushroom cell

Figure 6. 3D extruded view of the mushroom cell for (a) x-y plane with 10 nm out-of-plane depth (zz direction) and (b) out of mushroom-plane (yy–zz) with 4 nm out-of-plane depth (xx direction) 2D simulations. (a) shows the current density inside the amorphous region.

Reset, Set and read pulse voltages and current

Figure 7. Reset, Set and read pulses voltages (VappliedV_{\mathrm{applied}}, VgV_g and VcellV_{\mathrm{cell}}) and current. Set pulse has longer fall time compared to Reset pulse. PCM cells go through higher temperature in Reset than Set operation. Reading is performed with Vg=2V_g = 2 V to keep the n-MOSFET at low resistance, and Vapplied=0.1V_{\mathrm{applied}} = 0.1 V not to disturb the state of the cell.

Reset Operation

Reset operation results in different amorphous volumes and resistances for different access transistor (nMOSFET) gate pulse amplitudes (VgV_g) at constant PCM electrode bias (VappliedV_{\mathrm{applied}}) (Figure 1). Increasing VgV_g of the nMOSFET leads to increased Reset current, IresetI_{\mathrm{reset}}, which increases the size of the amorphous volume and memory window (Figure 8, Figure 9). RresetR_{\mathrm{reset}} significantly increases if an amorphous region completely covers the bottom contact. The increase in RresetR_{\mathrm{reset}} is relatively mild for higher IresetI_{\mathrm{reset}} beyond this level

Cell resistance after Reset operation versus Reset current

Figure 9. Cell resistance after Reset operation versus Reset current for top and bottom bias polarities. (a-f) show Reset cell crystallinity profiles (cropped figures) for different Reset currents.

(Figure 9). Similar saturating behavior is observed for both top and bottom bias circuits with a significant difference in the required IresetI_{\mathrm{reset}}, and Reset power and energy, PresetP_{\mathrm{reset}} and EresetE_{\mathrm{reset}}, due to thermoelectric effects (Figure 10, Figure 11). The top-bias condition successfully resets at lower currents [27, 28, 29] (Figure 10). Higher power and energy can be delivered to the cell by increasing IresetI_{\mathrm{reset}} (Figure 10). However, over-reset is not desirable since power consumption and required access device size (limiting maximum IresetI_{\mathrm{reset}}) are major concerns for PCM design.

PCM cell peak Reset voltage, power, and energy versus Reset current

Figure 10. PCM cell peak Reset voltage and power and Reset energy versus Reset current. It shows top bias polarity requires ∼2×\sim2\times less voltage and current than the opposite polarity to achieve same RresetR_{\mathrm{reset}} (∼48 MΩ\sim48\ \mathrm{M}\Omega).

Reset resistance versus peak Reset power and Reset energy

Figure 11. RresetR_{\mathrm{reset}} versus peak Reset power and Reset energy for both polarities. Top bias Reset requires ∼3.5×\sim3.5\times less power and energy than the bottom bias polarity to achieve same RresetR_{\mathrm{reset}} (∼48 MΩ\sim48\ \mathrm{M}\Omega).

The thermoelectric term in Eq. 9, ∇⋅(JST)\nabla\cdot(\mathbf{J}ST), which can be written in terms of its electron and hole contributions as ∇⋅(JST)=∇⋅(JnSnT)+∇⋅(JpSpT)\nabla\cdot(\mathbf{J}ST)=\nabla\cdot(\mathbf{J}_{n}S_{n}T)+\nabla\cdot(\mathbf{J}_{p}S_{p}T), is very significant for Reset and Set operations due to (i) asymmetric cell structure, (ii) extreme current density and thermal gradients, and (iii) substantial generation and annihilation rate of electrons and holes at (or close-proximity to) the material and solid-liquid interfaces [30]. Fcc-GST and TiN are degenerate p-type and n-type semiconductors. The interface of these two materials is expected to experience substantial Peltier heating or cooling due to carrier recombination or generation, depending on the current polarity. Conduction in a-GST is expected to be due to free electrons due to band alignment and the inability to inject holes from the TiN or fcc-GST into the valance band of a-GST. Therefore, the TiN/a-GST interface is expected to have very small Peltier heating or cooling and the a-GST/fcc-GST interface is expected to have large Peltier heating or cooling. Molten GST is a very narrow band-gap semiconductor with

Cell resistance after Set operation versus Set current

Figure 13. Cell Resistance after Set operation versus Set current for top and bottom bias polarities. (a-f) Set cell crystalline orientation and peak temperature profiles for different Set currents. The cell was previously Reset with top bias polarity with 64 μA64\ \mu\mathrm{A}. Wheater=4 nmW_{\mathrm{heater}}=4\ \mathrm{nm}.

very large electron and hole concentrations (nn, pp). We estimate nliquid=pliquid∼1×1022 cm−3n_{\mathrm{liquid}}=p_{\mathrm{liquid}}\sim1\times10^{22}\ \mathrm{cm}^{-3} from the latent heat of fusion: latent heat of fusion / pair = kinetic energy of the electron + kinetic energy of the hole + band gap at Tmelt=3kT/2+3kT/2+3kT/2T_{\mathrm{melt}}=3kT/2+3kT/2+3kT/2.

Set Operation

Once the PCM cell is Reset, a subsequent Set pulse is applied to crystallize the GST (Fig. 7). IsetI_{\mathrm{set}} is controlled by VgV_g. The simulated cells, which start with identical reset condition (top-bias), can be set with 5 μA<Iset<64 μA5\ \mu\mathrm{A} < I_{\mathrm{set}} < 64\ \mu\mathrm{A} for top-bias and 5 μA<Iset<132 μA5\ \mu\mathrm{A} < I_{\mathrm{set}} < 132\ \mu\mathrm{A} for bottom-bias conditions Fig. 13. Excessive IsetI_{\mathrm{set}} results in melting and re-amorphization of the programming region. The bottom bias polarity has larger range for Set operation due to thermoelectric cooling at the interface between fcc-GST and the bottom TiN contact. The location of the dominant filament is a significant function of the thermal boundary conditions and the thermal boundary resistance (TBR) [31, 32, 33] between TiN and GST. Crystallization starts from one side of the mushroom, close to the bottom GST/SiO2_2 interface. If the simulations are conducted without TBR at the material interfaces, wider mushrooms are formed and the crystallization path depends on both the polarity and the shape of the mushroom Figure 12.

Crystal orientation angles and current density profiles for top and bottom polarity Set and Reset

Figure 12. Crystal orientation angles for fcc-GST and current density profiles for a-GST during top and bottom polarity Set of cells Reset with top and bottom polarity, without accounting for thermal boundary resistances.

Cell structures and resistance measurements for different bias conditions

Figure 13 (PDF p. 6).

Orthogonal-Plane (y–z) Simulations

We performed 2D simulations on the (y(y–z)z) plane in Figure 6(b), which is orthogonal to the mushroom (x(x–y)y) plane, and reduced the Reset (∼1\sim1 ns) and Set (∼4\sim4 ns) pulse durations to observe the role of filamentary conduction and variability along the depth of the cell Figure 15. The initial Reset pulse resulted in amorphization of a ∼35\sim35 nm section of the 50 nm depth of the cell, which is not a successful Reset. The second Reset pulse resulted in amorphization of the crystalline region that still bridged the two contacts, successfully resetting the cell. The subsequent Set and Reset pulses, completed in 5 ns and 4 ns, respectively, including the time to cool down to room temperature, form and break a bridge that is ∼10\sim10 nm in depth, leading to a clear (>100×> 100\times) memory window Figure 15. This small region, which is approximately 4 nm ×\times 10 nm ×\times 10 nm (x(x–yy–z)z), functions as the programming volume rather than the full semi-cylindrical amorphized mushroom region (∼10\sim10 nm ×\times 10nmnm\times5050 nm). Hence, 2D mushroom simulations with out-of-plane depth of 10 nm Figure 6(a) are expected to capture the operation dynamics of cells with larger out-of-plane depth.

Cell resistances after Reset and Set operation for 18 cycles

Figure 15. Cell resistances after Reset (RresetR_{\mathrm{reset}}) and Set (RsetR_{\mathrm{set}}) operation for 18 cycles (shown in Figure 14). The first Reset/Set cycle does not switch the cell but leads to successful switching starting from the next cycle. Rreset/RsetR_{\mathrm{reset}}/R_{\mathrm{set}} cycle-to-cycle variability is due to randomness in phase change dynamics and filamentary transport in a-GST.

Voltage and current waveforms for out-of-plane simulations

Figure 14. Shorter Reset (∼1\sim1 ns) and Set pulses (∼4\sim4 ns) voltage and current waveforms used for the out-of-plane simulations. VappliedV_{\mathrm{applied}} is kept constant, and only VgV_g is varied for Set/Reset operations.

Scaling of the out-of-plane depth is not expected to reduce the required IresetI_{\mathrm{reset}} and IsetI_{\mathrm{set}}, unless scaled below 10 nm, but can improve cell-to-cell and cycle-to-cycle variability. Larger out-of-plane depths can provide increased reliability as the cells would be more immune to over-reset: if the programming region is over-reset, current can still be flown at a different location to continue cycling of the cell Figure 15. While this increases variability, it will also increase the lifetime and cycling endurance of the cells.

Conclusions

We investigated thermoelectric effects and filamentary transport in mushroom phase change memory cells during Reset and Set operations using an electrothermal finite element framework that accounts for temperature-dependent transport parameters, phase change dynamics, and stochastic activation energy distribution in the amorphous phase. We observe a 2×2\times reduction in reset current and 3.5×3.5\times reduction in energy consumption and peak Reset power with top-polarity bias condition due to thermoelectric effects. Our model predicts a significant change in required Set current as a function of mushroom shape and size, but no clear advantage for either bias polarity. Thermal boundary conditions and especially thermal boundary resistances at the GST/TiN interface at the bottom contact play a significant role in Reset and Set dynamics, mushroom shape,

current and peak power requirements. Filamentary conduction in a-GST during Set operation initiates local heating and phase change and determines the location, shape and volume of the programming region, which may be significantly smaller than the semi-cylindrical mushroom region. In the case of relatively wide amorphized mushrooms, we observe crystallization starting from the top of the mushroom for the top-bias Set and from the side of the mushroom for bottom-bias Set. Out-of-plane simulations capture cycle-to-cycle variability and show that cells with larger out-of-plane depth are expected to show increased variability but also increased lifetime and cycling endurance. These findings highlight the central role of filamentary transport and thermoelectric effects in PCM switching dynamics and energy consumption and contribute to the broader effort of advancing PCM toward scalable, low-power, and reliable memory technologies for future computing applications.

Acknowledgment

The authors thank F. Dirisaglik for resistivity measurements, S. Muneer for activation energy extraction, and Z. Woods for constructing the phase change physics model.

References

References

  1. [1]W. Zhang, R. Mazzarello, M. Wuttig, and E. Ma, “Designing crystallization in phase-change materials for universal memory and neuro-inspired computing,” Nature Reviews Materials, vol. 4, pp. 150–168, 2019.DOI
  2. [2]G. W. Burr, M. J. BrightSky, A. Sebastian, H.-Y. Cheng, J.-Y. Wu, S. Kim, N. E. Sosa, N. Papandreou, H.-L. Lung, H. Pozidis et al., “Recent progress in phase-change memory technology,” IEEE Journal on Emerging and Selected Topics in Circuits and Systems, vol. 6, no. 2, pp. 146–162, 2016.
  3. [3]A. Redaelli, E. Petroni, and R. Annunziata, “Material and process engineering challenges in ge-rich gst for embedded pcm,” Materials Science in Semiconductor Processing, vol. 137, p. 106184, 2022.DOI
  4. [4]P. Cappelletti, R. Annunziata, F. Arnaud, F. Disegni, A. Maurelli, and P. Zuliani, “Phase change memory for automotive grade embedded NVM applications,” Journal of Physics D: Applied Physics, vol. 53, no. 19, p. 193002, 2020.DOI
  5. [5]A. Faraclas, N. Williams, A. Gokirmak, and H. Silva, “Modeling of SET and reset operations of phase-change memory cells,” IEEE Electron Device Letters, vol. 32, no. 12, pp. 1737–1739, 2011.DOI
  6. [6]M. S. B. Hafiz, H. Silva, and A. Gokirmak, “Modeling filamentary conduction in reset phase change memory devices,” physica status solidi (RRL) – Rapid Research Letters, p. 2400416, 2025.
  7. [7]S. Raoux, F. Xiong, M. Wuttig, and E. Pop, “Phase change materials and phase change memory,” MRS Bulletin, vol. 39, no. 8, pp. 703–710, 2014.DOI
  8. [8]S. Raoux, G. W. Burr, M. J. Breitwisch, C. T. Rettner, Y.-C. Chen, R. M. Shelby, M. Salinga, D. Krebs, S.-H. Chen, H.-L. Lung, and C. H. Lam, “Phase-change random access memory: A scalable technology,” IBM Journal of Research and Development, vol. 52, no. 4.5, pp. 465–479, 2008.DOI
  9. [9]Y. Yu and M. Skowronski, “Observation and modelling of homogeneous nucleation in ge₂sb₂te₅ mushroom cells during set operation,” Journal of Materials Chemistry C, vol. 11, no. 11, pp. 3848–3854, 2023.
  10. [10]H.-Y. Cheng, F. Carta, W.-C. Chien, H.-L. Lung, and M. J. BrightSky, “3d cross-point phase-change memory for storage-class memory,” Journal of Physics D: Applied Physics, vol. 52, no. 47, p. 473002, 2019.DOI
  11. [11]D. Elmini, Phase Change Memory: Device Physics, Reliability and Applications. Cham: Springer, 2018.
  12. [12]D. Elmini, “Threshold switching mechanism by high-field energy gain in the hopping transport of chalcogenide glasses,” Physical Review B, vol. 78, no. 3, p. 035308, 2008.
  13. [13]F. Pan, S. Gao, C. Chen, C. Song, and F. Zeng, “Recent progress in resistive random access memories: Materials, switching mechanisms, and performance,” Materials Science and Engineering: R: Reports, vol. 83, pp. 1–59, 2014.DOI
  14. [14]M. Nardone, M. Simon, I. V. Karpov, and V. G. Karpov, “Electrical conduction in chalcogenide glasses of phase change memory,” Journal of Applied Physics, vol. 112, no. 7, p. 071101, 2012.DOI
  15. [15]F. Dirisaglik, G. Bakan, Z. Jurado, S. Muneer, M. Akbulut, J. Rarey, L. Sullivan, M. Wennberg, A. King, L. Zhang et al., “High speed, high temperature electrical characterization of phase change materials: metastable phases, crystallization dynamics, and resistance drift,” Nanoscale, vol. 7, no. 40, pp. 16625–16630, 2015.DOI
  16. [16]K. Cil et al., “Electrical resistivity of liquid Ge2Sb2Te5 based on thin-film and nanoscale device measurements,” IEEE Transactions on Electron Devices, vol. 60, no. 1, pp. 433–437, 2013.DOI
  17. [17]S. Muneer, J. Scoggin, F. Dirisaglik, L. Adnane, A. Cywar, G. Bakan, K. Cil, C. Lam, H. Silva, and A. Gokirmak, “Activation energy of metastable amorphous Ge2Sb2Te5 from room temperature to melt,” AIP Advances, vol. 8, no. 6, p. 065314, 2018.arxiv.org/abs/1805.04054
  18. [18]R. S. Khan, A. H. Talukder, F. Dirisaglik, H. Silva, and A. Gokirmak, “Accelerating and stopping resistance drift in phase change memory cells via high electric field stress,” arXiv preprint arXiv:2002.12487, 2020.arxiv.org/abs/2002.12487
  19. [19]M. Tashfiq, B. Kashem, R. S. Khan, A. H. Talukder, F. Dirisaglik, and A. Gokirmak, “Stopping resistance drift in phase change memory cells and analysis of charge transport in stable amorphous Ge2Sb2Te5,” arXiv preprint arXiv:2210.14035, 2022.
  20. [20]J. L. Lin and S. C. Lee, “Amorphous silicon thin film transistors,” Journal of the Chinese Institute of Engineers, vol. 18, no. 4, pp. 451–460, 1995.DOI
  21. [21]J. Becker, E. Fretwurst, and R. Klanner, “Measurements of charge carrier mobilities and drift velocity saturation in bulk silicon of <111> and <100> crystal orientation at high electric fields,” Solid-State Electronics, vol. 56, no. 1, pp. 104–110, 2010.DOI
  22. [22]L. Adnane, A. Gokirmak, and H. Silva, “Electrical measurement setup for phase-change materials,” Review of Scientific Instruments, vol. 87, p. 075117, 2016.
  23. [23]Z. Woods and A. Gokirmak, “Modeling of phase-change memory: Nucleation, growth, and amorphization dynamics during SET and reset,” IEEE Transactions on Electron Devices, vol. 64, no. 11, pp. 4466–4471, 2017.DOI
  24. [24]J. Lee, E. Bozorg-Grayeli, S. Kim, M. Asheghi, H.-S. P. Wong, and K. E. Goodson, “Phonon and electron transport through Ge2Sb2Te5 films and interfaces bounded by metals,” Applied Physics Letters, vol. 102, no. 19, p. 191911, 2013.DOI
  25. [25]J. Scoggin, R. S. Khan, H. Silva, and A. Gokirmak, “Modeling and impacts of the latent heat of phase change and specific heat for phase change materials,” Applied Physics Letters, vol. 112, no. 19, p. 193502, 2018.DOI
  26. [26]J. Scoggin, Z. Woods, H. Silva, and A. Gokirmak, “Modeling heterogeneous melting in phase change memory devices,” Applied Physics Letters, vol. 114, no. 4, p. 043502, 2019.DOI
  27. [27]N. Cicchini, M. Laudato, A. Leone, P. Fantini, A. L. Lacaita, and D. Ielmini, “Impact of thermoelectric effects on phase change memory characteristics,” IEEE Transactions on Electron Devices, vol. 62, no. 10, pp. 3264–3271, 2015.
  28. [28]A. Faraclas, G. Bakan, L. Adnane, F. Dirisaglik, N. E. Williams, A. Gokirmak, and H. Silva, “Modeling of thermoelectric effects in phase change memory cells,” IEEE Transactions on Electron Devices, vol. 61, no. 2, pp. 372–378, 2014.DOI
  29. [29]M. T. B. Kashem, J. Scoggin, H. Silva, and A. Gokirmak, “(digital presentation) finite element modeling of thermoelectric effects in phase change memory cells,” ECS Transactions, vol. 108, no. 1, pp. 3–15, 2022.
  30. [30]G. Bakan, N. Khan, H. Silva, and A. Gokirmak, “High-temperature thermoelectric transport at small scales: Thermal generation, transport and recombination of minority carriers,” Scientific Reports, vol. 3, no. 1, p. 2724, 2013.DOI
  31. [31]J. Lee, E. Bozorg-Grayeli, S. Kim, M. Asheghi, H.-S. P. Wong, and K. E. Goodson, “Phonon and electron transport through Ge2Sb2Te5 films and interfaces bounded by metals,” Applied Physics Letters, vol. 102, no. 19, p. 191911, 2013.DOI
  32. [32]J. Lee, J. P. Reif enberger, E. Bozorg-Grayeli, L. Hom, Z. Li, S. B. Kim, M. Asheghi, H.-S. P. Wong, and K. E. Goodson, “Thermal conductivity of phase-change memory materials from cryogenic temperatures to 400 k,” in 12th IEEE Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems, 2010, pp. 1–8.
  33. [33]J. P. Reifenberg, K.-W. Chang, M. A. Panzer, S. Kim, J. A. Rowlette, M. Asheghi, H.-S. P. Wong, and K. E. Goodson, “Thermal boundary resistance measurements for phase-change memory devices,” IEEE Electron Device Letters, vol. 31, no. 1, pp. 56–58, 2010.

Paper details

vlnc:W_6btfehbw
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents