The accurate determination of the effective electrical channel length () of MOSFETs is of crucial importance for device and circuit characterization. As a dominant MOSFET device parameter, variation effect on circuit performance and functionality has to be accounted for during circuit design. With the magnitude of intra-die polysilicon length variation becoming a significant fraction of inter-die and inter-wafer variation, these variation can no longer be neglected as it was commonly done in the past (e.g. [111]). Indeed, the transistors in a circuit could have different value. For large die size, sub-half micron ULSI microprocessor chips, characterizing and understanding variation within the die is key to improving simulation predictability as well as enhancing parameteric circuit performance and process control. Furthermore, adjusting the circuit design methodology to deal with this issue is a major challenge facing circuit designers. A first step in this direction involves the accurate characterization of these variations.
Classical extraction techniques [104][103][19] cannot be used for the characterization of intra-die variation in submicron technology. These methods require the use of two or more different length devices, and are based on the following relationship between and the polysilicon gate length :
where it is assumed that is constant for all devices and that is equal to the drawn gate length. For sub-half micron CMOS technology, these assumptions are violated. Whereas Scanning or Transmission Electron Microscopy (SEM or TEM) can be used to accurately determine , the destructive nature of these techniques, and the need for multiple devices limit their applicability for intradie measurements.
In [42] a simple and accurate gate length extraction method valid down to device lengths of is described. It is based on the fact that the gate-to-source/drain capacitance () in the inversion region depends only on oxide thickness, geometrical device structure, device gate width and length, and polysilicon doping. The oxide thickness can accurately be determined using one accumulation capacitance measurement on a long channel device [87]. This value is not expected to change significantly for other device on the same die. The effects of variation in the device structure can also be neglected [42]. Thus the inversion capacitance of devices with large known width is mainly a function of the polysilicon gate length and the average polysilicon doping . These values can thus be extracted by matching experimental and MINIMOS simulated values with the device biased in the inversion region using nonlinear least-squares optimization. The MINIMOS simulation takes into account quantum mechanical and polysilicon depletion effects [87][33]. The method is non-destructive and its accuracy is immune to process variation which makes it applicable to intradie characterization. It is noted that the method is not sensitive to the characteristics doping profile provided the device is biased in the inversion regime. However, with an accurate doping profile and value, and can be extracted using data from a single device as opposed to the multiple devices required for the other approaches.
In the following, an extension to the extraction technique that bypasses the use of nonlinear least squares optimization is described. With this modification, the method becomes computationally efficient and could therefore be incorporated as part of routine measurement procedures.