HP41XX - Datasheet Archive
Spreading Resistance. S D Connor, S J Harrington, A J Manson, S Nigrin, S Thomas, M C Wilson Zarlink Semiconductor Ltd, Cheney
Automated On-Wafer Measurement of Noise Figure and Base Spreading Resistance. S D Connor, S J Harrington, A J Manson, S Nigrin, S Thomas, M C Wilson Zarlink Semiconductor Ltd, Cheney Manor, SN2-2QW England. Email: firstname.lastname@example.org Abstract We present here details of a novel Rapid Modeling System (RMS). In particular the integration of automatic Noise Figure measurements (midband and high frequency) is described along with the software necessary to deliver statistical noise data and an alternative to `S' parameter Base Resistance extraction for noise sensitive design modeling. Introduction The Rapid Modeling System (RMS) is capable of measuring many devices on wafer and rapidly extracting both parametric and measurement data. The system is unique in two areas. First, it offers an almost open framework and general data management system into which new measurement techniques can be easily integrated. Second, it utilizes extensively developed ICCAP routines and SAS programs to facilitate either the measurement of single parameters or complete automatic model generation. The system can probe many thousands of devices per week and can be used to generate statistically valid models for Bipolar and MOS devices. Until recently all on-wafer noise measurements covering DC  to over 10GHz had been made manually. Noise measurements are of particular use to designers who need Noise Figure data for the design of cellular phone Ics, with the frequency of range 900 MHz to 1.9 GHz of specific interest. At the lower frequencies (determined by the device characteristics) parametric data can also be obtained from Noise Figure measurements, in particular information about the base spreading resistance parameters `Rbb, Rbm and Irb' . For these applications it is sensible to replace the usual `S' parameter derived Rb/Rbm values with the noise derived alternatives to facilitate better modeling of performance in this critical area. For these reasons it was decided that noise figure measurements would be automated and integrated into the Rapid Modeling System. This provides us with statistically valid noise profile data and an alternative method of assessing noise and Rbb spreads. The Rapid Modeling System A conceptual diagram of the Rapid Modeling System is shown in Figure 1. The whole system is initialized, configured and controlled by the extensive use of custom UNIX scripts running on Sparc Ultra workstations under Solaris 2.7. Environmental variables are set pertaining to process type, device structure and measurement type required. These variables are then used to start an ICCAP session with custom macros and transforms for the particular measurement required. The wafer is then autoprobed according to a co-ordinate set and device list, so each named device is tested at each given wafer map coordinate. Figure 1 RMS Hardware Configuration The system reads data files that contain catastrophic failure limits particular to the process, measurement and the device or devices being probed. After each instance of a device has been probed raw ASCII data is written (appended) until all sites have been measured. If a device falls outside of the catastrophic failure limits then special error codes are set. Control is then passed back to the UNIX environment, which initializes a remote copy of the SAS software. The SAS session thus invoked employs custom macros to filter the data and convert it into SAS data sets suitable for further processing or display. The hardware systems are based around three Cascade Summit Autoprobers (two 12000 machines and one 10000) which are usually dedicated to AC, DC and capacitance measurements (Figure 2).The matching system although `stand alone' , uses the same data format as the rest of the autoprobe system and interfaces at the `Data Warehouse' level. This system has been described elsewhere  Figure 2 RMS Hardware Configuration Test Structures The test chip was based on a Zarlink Semiconductor `standard' RF probe cell used extensively for RF probe measurements. Figure 3 shows a typical common emitter cell which is probed by two Cascade-Microtech coplanar GSG (Ground Signal Ground) 50 GHz probes. Sufficient room was available on the chip to place a five by four array of these structures. Unlike the standard RF cell all device placements were single instances with no parallel structures allowed .Previously, the only structures available were non (noise) dedicated RF cells with some devices only being represented within paralleled form . One cell of each chip was also reserved for a `through' calibration structure. Figure 4 Noise Figure Residuals. It can be seen that the trend is a fairly monotonic increase in measurement residual as the probe contact degrades with each successive touch down. Without recalibration the residual noise figure quickly becomes unacceptable Our acceptance criteria is that after calibration the noise figure should be less than +/- 0.1 dB with the probes on the calibration structure both before and immediately after a measurement has been made.If this is not the case then the mea surement is discarded and the system is re-calibrated using the local `through' structure. The +/-0.1dB limit is somewhat arbitrary and can be revised depending on the primary focus of the Noise Figure extraction. If the Rbb parameters are of primary interest then this corresponds to an extraction error of around +/- 3 . Whether this is of significance depends upon the size of the device. Measurements are also discarded if the insertion gain of the device is less than -3dB Hardware Configuration The hardware configuration for the noise data extraction consists of the following: i) HP 8970 B Noise Figure Meter Figure 3 RF Cell Structure Calibration Structures Automated Noise Figure measurement requires that careful consideration be given to system stability and especially calibration drift due to probe contact degradation. Previous work has shown that with this type of probe a local calibration structure is essential. Figure 4 shows the cumulative noise figure error (residual) after 30 successive probe contacts with only an initial calibration. ii) Maury Microwave Frequency Extender iii) HP 346B Noise Source iv) Bias System (HP41XX HP41XX) and suitable bias `T's v) HP 8341 Frequency Synthesizer Care must be taken to ensure that the 3dB points of the bias `T' system is adequate for the desired measurement frequency range. Figure 5 RMS AC/Noise Hardware Figure 5 details the complete hardware configuration with the Noise extender unit in place. With this system Noise Figure measurements can be made from 10MHz to 1.6 GHz without the extender and 1.6GHz through 18Ghz with the extender in place. The coverage is not quite continuous as the 3dB points of the Noise Figure Meter (upper) and the Frequency Extender (lower) place a `notch' in the response Measurement Integration All RMS autoprobe applications adhere to a standard form although this is a loose constraint. The system records the following data for each site measured. · Site id and device type · Gummel Plot ( Ic Ib for each bias point) · Noise Figure and Insertion Gain · Noise Residual before and after measurement. The measurement frequency was chosen to be 900 M Hz and 12 bias points from 0.5 uA to 1mA were set. Forty sites were autoprobed in total. Figure 7 shows a SAS distribution of the minimum noise figure after filter and Figure 8 illustrates a regenerated Noise Figure plot with statistics. The software can then, with constraints extract the Rb data from the above using the relationship: Vce and Frequency · Figure 6 Typical IG and Nf Plot from IC-CAP Figure 6 shows a typical Noise Figure and Insertion Gain plot for a single site measurement. Although a general wafer mapping and graph plotting facility is available within the system it was decided to add a further SAS program to deal specifically with noise measurement data. Example Results The following results are taken from two wafers of Zarlink Semiconductor's advanced bipolar technology `HJ'. HJ is a complementary RF process with a self aligned base-emitter structure and cut off frequencies in excess of 25 GHz. The device measured was a large multiple emitter transistor (designated n1dcy4) known to have a minimum noise figure of around 2.5 dB. rb (rb + Rs ) qIe qIe rb + Rs + re + + Rs 2kTRs 2kTRs 2 Nf ( dB ) = 10 log 1 + 2 This data can then be used to extract and optimize the SPICE parameters Rb, Rbm and Irb. The extraction and optimization is coded directly into SAS. This bypass of the IC-CAP optimization significantly decreases the processing time associated with this operation. Figure 9 shows the raw data pre optimization with the initial simulated data being generated from `S' parameter derived data. Also shown is the simulated curve after optimization with the new parameters displayed. We would generally expect that noise derived Rb data would be somewhat higher -typically by a factor 1.5 to 2- than for `S'parameter data as seen in Figure 10. Finally for a `closed' loop we can feed the extracted Rb parameters back into our simulator and attempt to regenerate the Noise Profile. This is shown in Figure 10 where we see good agreement between the regenerated (simulated) noise profile and the original data. Figure 9 Optimized Rb parameters Figure 7 Minimum Noise Figure Data Figure 10 Measured Noise Profile Against Simulation References  Agilent EEsof EDA IC-CAP. Figure 8 Noise Profile With Statistics Conclusions We have presented a model of the autoprobe system The open architecture of the system has facilitated the easy inclusion of a time intensive measurement, which had been until this time strictly manual. SAS code has been generated which will will be integrated into the general graphing utility (macro). In particular we have described the implementation within the system of a facility capable of measuring and presenting both mid-band and high frequency noise data. Finally we have presented a method of automatically extracting the Base Spreading Resistance (Rb) from the noise data and optimizing SPICE parameters within SAS.  SAS Software The SAS Institute.  S.D. Connor; D. Evanson; Proceedings of IEEE International Conference on Microelectronic Test Structures, pp33-38, (March 25-28, 1996).  S.D. Connor, Proceedings of IEEE International Conference on Microelectronic Test Structures, pp43-48, (March 17-20, 1997).  S.D. Connor, Proceedings of IEEE International Conference on Microelectronic Test Structures, pp33-38, (March 15-18, 1998).