Semi-empirical neural network modeling of metal-organic chemical vapor deposition

Abstract

Metal-organic chemical vapor deposition (MOCVD)is an important technique for growing thin films with various applications in electronics and optics. The development of accurate and efficient MOCVD process models is therefore desirable, since such models can be instrumental in improving process control in a manufacturing environment. This paper presents a semiempirical MOCVD model based on "hybrid" neural networks. The model is constructed by characterizing the MOCVD of titanium dioxide (TiO 2 ) films through the measurement of deposition rate over a range of deposition conditions by a statistically designed experiment in which susceptor and source temperature, flow rate of the carrier gas for the precursor and chamber pressure are varied. A modified backpropagation neural network is then trained on the experimental data to determine the value of the adjustable parameters in an analytical expression for the TiO2 deposition rate. In so doing, a general purpose methodology for deriving semi-empirical neural process models which take into account prior knowledge of the underlying process physics is developed.

Hybrid Neural Network Modeling of Anion Exchange at the Interfaces of Mixed Anion III–V Heterostructures Grown by Molecular Beam Epitaxy

“…The function of each neuron is to compute the weighted sum of its inputs and determine its activation level using a nonlinear transfer function. This can be expressed analytically as (9) where is the synaptic weight, is the input, and is the output corresponding to the th neuron in the th layer. The nonlinear activation function used to determine the output of each neuron enables neural networks to generalize with a degree of freedom not available in statistical techniques [8].…”

Neural network based uniformity profile control of linear chemical-mechanical planarization

“…• Step 4: At th run, update the feedforward neural network process model by adaptation law (8) with the new data set. • Step 5: After the feedforward process model has been updated, use adaptation law (12) to calculate the new weights for the uniformity controller based on the estimate of removal rate profile and then go to step 3.…”

“…For the neural network based controller, we can obtain the optimized recipe by learning law given by (12) as - (15) Some experiments using -were tested. We found that the process performance under recipe -was not satisfactory.…”

“…• Step 2: A neural network based uniformity controller is trained off-line by a BP algorithm by (12) through the use of the feedforward neural network model obtained in step 1. • Step 3: The trained neural network based uniformity controller is used to tune the air-bearing pressures and platen height at the th run for the CMP process.…”

“…In [10], a neural network model was introduced to study the plasma etching processes, and superior performance has been demonstrated compared with the SRM approach. Even though the neural network method has been applied to other semiconductor manufacturing processes such as CVD [11], [12], plasma and ion etchings [10], [13], [14], a small amount research has been performed for CMP processes. Lin and Liu [15] used an adaptive neuro-fuzzy interface system to analyze CMP process parameters on rotary CMP tools; however, very limited experiments and polishing parameters have been studied.…”