Thermal assistance has been shown to significantly reduce the required operation power for spin torque transfer magnetic random access memory (STT-MRAM). Proposed heating methods include modified material stack compositions that result in increased self-heating or external heat sources. In this work we analyze the self-heating process of a standard perpendicular magnetic anisotropy STT-MRAM device through numerical simulations in order to understand the relative contributions of Joule, thermoelectric Peltier and Thomson, and tunneling junction heating. A 2D rotationally symmetric numerical model is used to solve the coupled electro-thermal equations including thermoelectric effects and heat absorbed or released at the tunneling junction. We compare self-heating for different common passivation materials, positive and negative electrical current polarity, and different device thermal anchoring and boundaries resistance configurations. The variations considered are found to result in significant differences in maximum temperatures reached. Average increases of 3 K, 10 K, and 100 K for different passivation materials, positive and negative polarity, and different thermal anchoring configurations, respectively, are observed. The highest temperatures, up to 424 K, are obtained for silicon dioxide as the passivation material, positive polarity, and low thermal anchoring with thermal boundary resistance configurations. Interestingly it is also found that due to the tunneling heat, Peltier effect, device geometry, and numerous interfacial layers around the magnetic tunnel junction (MTJ), most of the heat is dissipated on the lower potential side of the magnetic junction. This asymmetry in heating, which has also been observed experimentally, is important as thermally assisted switching requires heating of the free layer specifically and this will be significantly different for the two polarity operations, set and reset.
Keywords: FEM modeling; Joule heating; self-heating; spin torque transfer magnetic random access memory (STT-MRAM); thermoelectrics
Spin torque transfer magnetic random access memory (STT-MRAM), a type of non-volatile memory, functions through the resistance ratio between the parallel (ON) and anti-parallel (OFF) states of two ferromagnetic domains on either side of a thin insulating barrier. These ferromagnetic domains and the insulating barrier make up the magnetic tunnel junction (MTJ) of the STT-MRAM device. The device is switched between the ON and OFF states by passing a current of the appropriate direction through the MTJ allowing the rotational spin of tunneling carriers to “torque” the free magnetic domain into the desired alignment. STT-MRAM exhibits desirable endurance, speed, and scaling properties. The write process of STT-MRAM has been shown to be highly reactive to thermal processes [1-3]. Bi et al. showed that utilizing external heat sources to elevate the temperature of the device by 50 K could reduce power consumption by up to 4.8% . Bandiera et al. showed practical operation of STT-MRAM devices below the 22 nm node by enhanced self-heating, up to temperatures of 360 K to 523 K, through manipulation of the material stack . However, changing the material stack compromises the desired anisotropy levels and external heat sources impact device density. Thus, these methods for thermal assistance can be considered limited in their potential.
In this work we perform an electro-thermal analysis of the self-heating process for a standard perpendicular magnetic anisotropy STT-MRAM device (Figure 1) to understand the relative contributions of the different heat mechanisms involved and the effect of external device parameters such as passivation material, current polarity and contact configurations, and to determine how such design choices may enhance thermal assistance for STT-MRAM.
We have modeled the electro-thermal behavior of the device using the “COMSOL Multiphysics Finite-Element Modeling” software. Voltage, current, and temperature are obtained by solving the coupled current (Equation 1) and heat (Equation 2) equations:
where is the current density, σ is the electrical conductivity, is the electric field, is the external current source used to model the tunnel junction, ρ is the mass density, Cp is the heat capacity, T is the temperature, t is the time, k is the thermal conductivity, and Π is the net Peltier coefficient. Thermoelectric Peltier, and Thomson heat terms are included in the heat equation (Equation 2). Tunneling through the device is modeled using an external circuit that circumvents the thin insulating barrier with the resistance being characterized by experimental J–V curves obtained for the same junction from literature [4-7] (Figure 2).
This external circuit to model the tunnel junction is the only contribution of and requires the corresponding heat contributions to be added separately as external heat sources. Heat released or absorbed at the tunneling junction (Qt) is modeled using the probabilistic equation for hot tunneling carrier relaxation :
where is the magnitude of tunneling current density, Vt is the electric potential drop across the junction, λ is the inelastic scattering mean free path, and x is the stack position. The Peltier heat associated with the tunneling electrons is modeled using:
where is the tunneling current density, and S and T are the Peltier coefficients (S is the Seebeck coefficient, T is the temperature) on either side of the junction. The heat contributions from Equation 3 and Equation 4 are released or absorbed on the appropriate side of the tunneling barrier depending on the current polarity.
The MRAM device modeled (Figure 1) is composed of the following layers: contact | 0.5 Ta | 0.9 CoFeB (free) | 1 MgO | 0.5 Fe | 0.5 CoFeB (fixed) | 0.6 Ta | [0.25 Pd | 0.8 Co] × 10 | 4 Ru | 4 TaN | 20 Cu | 7 Ru | contact, where the numbers before the alloy composition of each layer represent the thickness of that layer in nanometers. The radius of the cylindrical material stack is 10 nm. The passivation layer surrounding the stack is 5 nm thick. In the material stack the CoFeB layers are the ferromagnetic domains of the MTJ, MgO is the thin insulating layer, and Ta/CoPd/Ru is the synthetic antiferromagnetic layer. The material stack modeled is from a device reported by Worledge et al. . The thermal stability factor of the free ferromagnetic layer of the device, derived from Equation 5 using materials properties from [8,9], is 38.7 kBT:
where Δperp is the perpendicular thermal stability factor, Ms is the saturation magnetization, is the perpendicular effective anisotropy field, V is the volume of the free layer, kB is the Boltzmann constant, and T is the local temperature.
where L is the Lorenz number. It is assumed the annealing temperature was sufficiently low so as to not cause crystallization  on the MgO–CoFeB interface, leaving the CoFeB completely amorphous. The Pd–Co layers are treated as an alloy due to the large number of sub-nanometer thick layers. The temperature-dependent materials properties of the PdCo alloy were derived using the behavior of similar alloys  transposed onto point data . The material properties of the various common passivation materials considered  (Table 1) are from the element library of COMSOL for SiO2 and Si3N4, and from literature for low-temperature plasma-enhanced chemical vapor deposition SiO2 [2,17] and Si3N4 . The “ON” and “OFF” states of the device are represented via interchangeable sets of materials properties for the CoFeB layers [4,13] (Figure 2, Figure 3) that are chosen based on the desired switching operation to be simulated (anti-parallel properties to simulate an OFF–ON switching operation, parallel properties for the converse operation, see inset in Figure 2).
Table 1: Properties of passivation materials.a
|material||relative permittivity||thermal conductivity (W/m·K)||heat capacity (J/kg·K)|
aProperties at room temperature are given. Values are from the “COMSOL” element library for SiO2 and Si3N4, (high-temperature growth or deposition, considered here as a limit case) and from literature for SiO2* [1,15] and Si3N4* . bThe asterisk indicates material properties for low-temperature plasma-enhanced chemical vapor deposition (PECVD) silicon dioxide or silicon nitride.
Analysis of the self-heating of the device is performed for different passivation materials (Table 1), positive and negative current polarity, and four contact configurations (Figure 4), using voltage pulses as shown in the inset of Figure 1. The bias voltage dependence of the TMR is not considered and the resistance values of the MTJ in the ON and OFF states, derived from Figure 2, are assumed to be constant during the switching pulses. A positive voltage is applied at point A and ground at point B in Figure 4 (positive polarity, to simulate ON–OFF switching) or the converse (negative polarity, to simulate OFF–ON switching). The thermal boundary conditions, represented as C in Figure 4, are set to 300 K at all times. The peak temperatures within the free CoFeB layer, fixed CoFeB layer, and for the whole device are recorded for each pulse as Tfree, Tfixed, and Tpeak, respectively.
The contact regions in Figure 1 are changed between sets of simulations in order to model the device in different thermal anchoring conditions. The four configurations used to model these contact regions are shown in Figure 4. To model a discrete STT-MRAM cell, the configurations I and II are simulated and referred to as “small contact” devices. To model a device in a memory array, the “large contact” configurations III and IV are used to emulate large thermal coupling . To model different heat paths through metal and passivation layers, the configurations with passivation–Cu contacts (I, III) are simulated with and without thermal boundary resistances (TBR) applied on the passivation–Cu interfaces. This TBR is modeled as a 1 nm thick virtual layer with a thermal conductivity of 0.041 W/m·K.
Simulation sets are performed for configurations II and IV in positive polarity, with pulse amplitudes ranging from 0.2 to 0.4 V, with all passivation materials considered in Table 1. Additional simulations are run for SiO2 passivation layer, which exhibits the most favorable contributions to self-heating (see Results and Conclusion), on configurations I to IV with pulse amplitude ranging from 0.2 to 0.6 V for both positive and negative polarity to further analyze the effects of thermal anchoring and current polarity.
Asymmetry of heating around the MTJ, shown by the simulation results in Figures 5–8, is a result of the asymmetric heat contributions from QPeltier and QTunneling and interfacial layers with high thermal resistance surrounding the thin insulating MgO layer. The asymmetry in heating increases with overall heating as shown below in Figure 7 and Figure 8 for configurations II and IV. Experimental observations of asymmetric heating in MRAM devices have been reported earlier by Gapihan et al. who predicted that the favorable current direction can result in a 10% reduction in the heating power density required for switching .
The temperature Tpeak of the small-contact configurations at expected operational voltages of STT-MRAM (0.3 V) range from 305 K for configuration II in positive polarity with Si3N4 passivation, to 424 K for configuration I with TBR in positive polarity with SiO2 passivation.
Large-contact configurations exhibit higher asymmetric tendencies and, on average, have a higher Tpeak by 11.8 K compared to observations for similar small-contact simulations (same passivation material, voltage applied, and polarity). This is the result of more pronounced geometric asymmetry and a larger heat capacity of the configurations III and IV. At typical operational voltages of STT-MRAM large-contact configurations have Tpeak ranging from 312 K for configuration IV in positive polarity with Si3N4 passivation to 326 K for configuration III in positive polarity with SiO2 passivation.
SiO2 passivation provides superior amplification to the self-heating of STT-MRAM with Tpeak that are, on average, 2.6 K higher than comparable simulations for the other considered passivation materials (low-temperature PECVD SiO2 and Si3N4, and Si3N4). This is observable in Figure 7 and Figure 8.
For the extended analysis of SiO2 passivation it is observed that small-contact configurations attain peak temperatures, on average, 111.2 K higher than their large-contact counterparts (i.e., the temperatures of configurations I and I-TBR are higher than those of configurations III and III-TBR). This is the result of the passivation material inhibiting heat flow (which is not present in configuration II), and smaller contact regions having lower heat capacity. Additionally, positive polarity yields, on average, 7.9 K higher Tpeak than negative polarity. This is due to reduced current in the negative polarity switching (device is initially in off state) and reduced distance to the thermal boundary for heat generated on the bottom of the MTJ. Assuming constant resistances for the two states – which is equivalent to assuming switching occurs only at the end of the pulses – together with temperature-dependent changes of electrical and thermal conductivity may result in a small underestimation or overestimation of the difference in heating between the positive and negative polarities. Nevertheless, the simulation results appear to be in reasonable agreement with the experimental observations of a reduction by ca. 10% in required power density for the positive polarity reported by Gapihan et al. . Expected temperature asymmetries for opposite current flow under the same resistance state should yield higher temperatures on the lower potential side of the MTJ, similar to the patterns observed in Figures 5–8. However, these cases are not simulated as the thermal effect is mainly of interest for current flow direction and resistance state combinations that result in device switching. Partial inclusion of the different self-heating mechanisms in the simulations shows the relative importance of each (Figure 9). The largest contributor to device heating, by a significant margin, is Qt. The Peltier and Joule heating in the device are minimal in comparison. As a result, heating is largely localized in the MTJ region as seen in Figure 5 and Figure 6.
Simulations were performed to observe the dependence of thermal behavior on the thickness of the passivation layer and no significant changes were observed between 5 and 20 nm thick layers (except at high voltages and low thermal anchoring where heat saturation occurs).
In this work, we analyzed the electro-thermal behavior of a standard STT-MRAM device for different passivation layers, current polarity and contact configurations and found that self-heating varies significantly and that, for some cases, it can effectively produce required temperatures of ca. 350 K  to about 523 K  for thermal assistance at expected operation voltages. The use of high-temperature PECVD SiO2 (with slightly higher thermal conductivity and heat capacity compared to low-temperature PECVD SiO2) as the passivation material results in significantly higher self-heating compared to the other common passivation materials (Si3N4 and low-temperature PECVD Si3N4 and SiO2). Positive polarity leads to a 10 K average temperature increase at negative polarity. Interestingly, the results also show that due to tunneling heat, Peltier effect, device geometry, and numerous interfacial layers around the MTJ, the heating on the lower potential side of the junction for a given polarity is higher. This asymmetry is important because thermal assistance requires heating of the free ferromagnetic domain, causing thermal destabilization. But the free domain only experiences significant heating when the device is in positive polarity. Thus care must be taken to ensure the free domain achieves a sufficiently high temperature independent of the overall device temperature.
The authors are thankful to Daniel Worledge for valuable discussions and suggestions that led to this work. A. Deschenes was supported by an REU supplement made to NSF Grant ECCS 1150960. S. Muneer, A. Gokirmak and H. Silva were partially supported by the U.S. Department of Energy through Basic Energy Sciences Award DE-SC005038. M. Akbulut was partially supported by the U.S. National Science Foundation through Award ECCE 1150960. The authors would also like to acknowledge Nadim Kanan, Nicholas Williams, and Zachary Woods for general discussions and helpful criticism.