Merits of Batch ALD

Abstract

An overview of the merits and applications of batch ALD in a vertical furnace will be presented. We address new material and process developments and throughput enhancement which are key factors for future high-volume manufacturing applications. We present ALD SbOx as a new material on the batch platform. For the workhorse ALD Al2O3 and TiN materials, experimental and simulation results demonstrate that a reduction in cycle time to <21s does not significantly compromise uniformity, resistivity and step coverage. Moreover, batch ALD offers process flexibility in the mitigation of non-idealities, such as growth inhibition, as will be discussed for thermal ALD AlN. The very low non-uniformities < 1% (1σ) for the latter process demonstrate the competitive film properties that can be achieved by batch ALD. With applications as diverse as metal gates in logic, trench capacitor electrodes, capacitor dielectrics, barrier layers and passivation films, batch ALD has firmly established itself at device manufacturers and foundry sites with significant prospects for emerging markets.

Material insights of HfO2-based integrated 1-transistor-1-resistor resistive random access memory devices processed by batch atomic layer deposition

“…The TiN/Ti/HfO 2 /TiN MIM structures, acting as the resistor, were fabricated by depositing HfO 2 at two different temperatures (150 °C and 300 °C) by employing the batch ALD approach (with 100 process-wafer loading capability) with a metal organic precursor (which is more suitable for the batch ALD process thanks to its liquid form22). Both electrical and material properties of devices were systematically studied and correlated.…”

Integrating Feedback Control and Run-to-Run Control in Multi-Wafer Thermal Atomic Layer Deposition of Thin Films

“…In this subsection, the CFD model result for the furnace ALD system is validated with experimental results and existing simulation models. The current models reported in [8] and in [9] utilized empirical knowledge and a rudimentary gas-phase CFD computation to analyze the transport time scale. Specifically, the time scale results were compared to the 2D asymmetric multiscale CFD model in this paper under similar operating conditions: a precursor inlet flow rate of 1.3 slm, a precursor mole fraction of 0.3, and a vacuum pump setting of 100 Pa and 266 Pa.…”

“…As shown in Figure 11, a series of control actions with a sampling time of 1 s were carried out according to the PI control scheme, which reduced the inlet feed mole fraction to 0.226 and drove the precursor partial pressure to the set-point and stabilized the process output after 18 s. Also, the controller successfully reduced 23% of the precursor usage in the first batch. With the updated feed mole fraction determined from the PI controller, the relationship between the inlet flow rate and the half-cycle time, τ, under the standard inlet flow rate, q, range was again determined to be: τ(q) = −0.11q + 322.16 (9) According to the reactor response under the updated feed mole fraction, a half-cycle time of 130 s was chosen to be the set-point for the R2R control, and a corresponding q sp is 1700 sccm. As shown in Figure 12, the diamond (red) data points represent the simulated half-cycle time under the existing pressure disturbance, and the downward (blue) and upward (green) triangular data points represent those under the R2R control schemes with a learning factor λ of 0.25 and 0.75, respectively.…”

“…The lack of understanding of the influence of the gas-phase profile on the wafer surface deposition is one of the most important problems [8]. In an experimental work and industrial pilot plant, the operating cost associated with the detailed tests in exploring the optimal operating conditions for the batch ALD systems, especially furnaces, is also very high due to the large physical scale and the long processing time [9]. Thus, experimental works have only been able to investigate a limited range of precursor selections and operating conditions.…”

“…[15] proposed a machine learning based data-driven model that largely cut down the computation time and preserved model fidelity. However, furnace ALD simulations were mainly still based on semi-empirical Langmuir isotherm, empirical relationships, and simplified CFD gas-phase simulation [8,9].…”