The energy of ions bombardmg the wafer is proportional to the potential difference between the plasma and the powered electrode in reactive ion etching systems. This work seeks to control the ion energy without altering the applied radio-frequency power or the chamber pressure since these variables are closely tied to other important quantities, such as reactive chemical species concentrations in the plasma and wafer etch uniformity. A variable resistor placed in parallel with the blocking capacitor allows the plasma self-bias voltage (Vbi,,) to be arbitrarily varied between its nominal value and zero. Optical emission spectroscopy for a CF, plasma reveals that the nominal plasma chemical concentrations do not change under this control method. The use of a Langmuir probe to measure the plasma potential shows that the ion energy changes by approximately one-half of the change in Vbi,,. The potential uses of this ion energy control technique to plasma self-bias voltage regulation, etch selectivity, and plasma cleaning of chamber walls are demonstrated. A potential drawback, namely, decreased plasma stability, is also indicated. ) unless CC License in place (see abstract). ecsdl.org/site/terms_use address. Redistribution subject to ECS terms of use (see 137.99.31.134 Downloaded on 2015-06-18 to IP

No abstract

The errors present in electrical measurements at frequencies and impedances relevant to plasma processing in the semiconductor industry are studied. A theoretical bound on calculated delivered power error as a function of measured electrical values is derived. The derivation shows that for constant measurement error, power error is a linear function of load impedance expressed in terms of voltage standing wave ratio. This bound is supported by experimental data taken with both a directional coupler and a voltage-and current-based probe. Linear and nonlinear model based sensing methods are implemented which reduce power error by a factor of five over a standard calibration. Published results are cited which indicate that the voltage standing wave ratio of typical plasma processing and experimental regimes is high enough to cause small measurement errors to result in large calculated power errors. Nomenclature electrical state: electrical state, no representation chosen load: a one port network with a specific impedance or reflection coefficient, no representation chosen probe: measured values taken at the probe's measurement ports srp: 'sensor reference plane' the reference plane inside the sensor where measurement actually takes place. See Section 6.1 lrp: 'load reference plane'-the reference plane at the point in the circuit where a load is connected to the probe. See Figure 6 x ref : electrical state at ref point in circuit, no representation chosen (V + , V ,) : electrical state represented as forward and reverse voltage waves (v, i) : electrical state represented as current and voltage '(var)-based' : electrical state measured as 'var ' var : actual value of var

No abstract

The sensitivity of a novel broad frequency band (1–2.25 GHz) radio frequency sensing system to plasma etching process conditions is demonstrated. This is accomplished by using the sensing system to estimate polysilicon etch rate in a Lam 9400 etch tool. A designed experiment varying physical and chemical reactive ion etching regimes was performed with five repetitions at each experimental point. A model relating broadband sensor response to etch rate was regressed using four repetitions of the data and validated on the fifth. Two representations of the broadband data were considered separately when regressing the models, with subset selection used in each case to choose the best predictor variables. In one representation, the sensor data was considered as a vector of 402 real numbers corresponding to magnitude and phase of reflection coefficient at each of 201 frequencies, resulting in an R2 of etch rate estimate of 0.997. In the other, the broadband response was parameterized on the basis of a multimodal cavity resonance model. The inferred parameters of natural frequency, quality factor and resistance were then used as the predictor variables for regression, resulting in an R2 of 0.962.