Calibrating a Typical Bipolar Process Flow


A common request from TCAD (Technology Computer Aided Design) users is advise on how to calibrate the process flow and device simulator to a particular technology. In this article, guidelines are presented for calibration of process and device simulations for a typical poly-emitter bipolar process. Such a process would be used to create a structure shown in Figure 1. Unlike a MOSFET process where calibration mostly takes place in the process simulation, calibration of a bipolar process will include parameters in both the process and device simulators. A simple procedure is presented here that highlights the most important parameters that require calibration and indicates the expected region of the gummel plot over which each variable has the most influence.

Figure 1. NPN bipolar transistor geometry and doping from ATHENA.



Calibration Method

The reader is presumed to be familiar with the basic operation of ATHENA and using appropriate models for each process step. For example, the incorrect use of diffusion models defined in the "method" statement, for example, will invalidate the rest of this following section.

Calibrating a bipolar process flow initially entails matching the two parameters, base current and collector current, to measured results throughout the full operating range of the device. By implication the current gain of the device (Ic/Ib) will also be matched. All of the following paragraphs refer to the standard plot of collector and base currents measured against the base-emitter voltage, Vbe, unless it is specifically stated otherwise. This standard I-V graph is usually referred to as the "Gummel Plot".

Another way of plotting the same information in a different format is a plot of current gain, hfe, versus the log of the collector current. This graph, however, is less clear as to which current is increasing or decreasing for each change of current gain. This graph is less useful for understanding variation of the collector and base currents.

The full operating range of a bipolar junction transistor (BJT) consists of three general regions defined by the current density injected into the base. These three operating regions are usually described as the low, medium and high current injection regimes. The medium injection region is the most important part of the curve to model correctly as this represents the typical operating condition of the BJT. Each of the three operating regions is dominated by a different physical phenomenon. Successful modeling of a BJT involves matching both the base and collector currents in each of the three general operating regions making a total of six areas for calibration. The derived parameter, hfe, is also a good parameter to monitor, since this is sensitive to errors in the ratio of collector to base current.


Identifying Five Calibration Targets

The following text suggests an approach and describes which of the five regions are affected by each change. The general technique is to calibrate the parameters that have the greatest effect on device performance in all regions first and then move to more subtle phenomenon that affect certain parts of the base or collector currents. In general, matching the collector current for all injection regions is less problematic than matching the base current at the extremes of the injection regions and so there are more sections on tailoring these parts of the curve. The text is therefore divided up into the following sections:

  1. Tuning Base and Collector Currents - All Regions
  2. Tuning the Base Current - All Regions
  3. Tuning the Collector Current - All Regions
  4. The Base Current Profile - Medium Injection
  5. The Base Current Profile - Low Injection

If the above order is followed, there should be a reasonable correlation between measured and simulated data. However, most of the tuning parameters have some degree of inter-dependency, the degree of which is also device design specific, so some degree of iteration of the tuning parameters is to be expected.

When tuning bipolar transistors, there is a greater emphasis in tuning device parameters, compared to optimizing MOSFETs where most tuning parameters are process related. A powerful combination is the tuning of a BiCMOS process where the MOSFET part of the process flow can be used to tune the process parameters while the bipolar part of the flow is used to tune the device simulator. This technique should yield a high degree of predictability in the results.

Tuning the process simulation parameters in Athena is mainly required to model affects such as the implantation induced defect enhanced diffusion, responsible for the "emitter push effect" which is essential to obtain the correct depth of the base-collector junction. The correct process modeling of the out diffusion of dopant from the poly-emitter into the mono-crystalline substrate is also critical to obtaining well matched I-V curves. Another critical process modeling area is the base implant as it is essential to match measured and modeled base resistance for correct modeling of the collector current. These and other issues are discussed in the sections below.


(i) Tuning Base and Collector Currents - All Regions

One important parameter to model the general level of base and collector currents is the device measurement temperature. The base and collector currents are strongly influenced by temperature changes as small as a few degrees centigrade. A significant effort should be made to determine the exact temperature of the device during measurements before calibration is attempted and this temperature should be input into the device simulator, ATLAS, in the MODELS statement using the TEMPERATURE=<> parameter. An increase in temperature will cause an increase in base and collector currents.


(ii) Tuning the Base Current - All Regions

A critical region for poly-emitter bipolar devices is the interface between the poly emitter and the mono-crystalline silicon. This region is difficult to process simulate directly as the interface between the poly-silicon emitter and single crystalline silicon usually consists of a thin, uneven and possibly non continuous film of oxide. The way this is simulated is by calibrating the overall affect of this interface, not with a process simulator but with the device simulator, ATLAS. The tuning parameter is the surface recombination velocity at this interface for electrons (VSURFN for PNP devices), or holes (VSURFP for NPN devices).

The surface recombination velocity parameter only affects the base current but affects the base current in all operating regions. It is therefore a powerful parameter to approximately match the base current and gain throughout the full operating range. In some cases the base current may be less affected in the very high and very low injection regions by changes in the surface recombination velocity adding some scope to fine tune the profile of the base current versus base-emitter voltage curve.

It is important to realize that the poly emitter MUST be defined as an electrode in order to be able to define the interfacial surface recombination velocity, VSURFN and VSURFP, using the CONTACT statement. This is in contrast to the MOSFET calibration text where is is strongly advised not to define the poly-gate as an electrode. Be sure not to get these two confused. The parameter that activates the recombination velocity is SURF.REC also in the contact statement. For an NPN BJT for example:


A lower value of recombination velocity, VSURFP will reduce the base current and increase the gain, hfe. The converse is also true.


(iii) Tuning the Collector Current - All Regions

The parameter which affects the collector current over the entire range is the intrinsic base resistance. The base resistance is primarily determined by the dose of the base implant(s). An increase in the base implant dose will decrease the intrinsic base resistance and decrease the collector current in all injection regions. In some cases, however, the collector current may be only slightly affected in the very high injection region giving scope for fine tuning the profile of collector current versus base-emitter voltage.

If the pinched or intrinsic base sheet resistance is a measured parameter, the simplest way to match measured and simulated data is to make slight changes to the base implant dose within the expected error actual implanted dose in conjunction with the error in percentage activation.

In some designs, where the base contact is close to the collector contact or the base contact is large, such as the substrate contact, the collector current may also be influenced in all current injection regions by specifying a surface recombination velocity at the base contact. For a typical design with a buried n+ collector and surface contacts, the surface recombination velocity at the base contact may have little effect on the collector current.


(iv) The Base Current Profile - Medium Injection

There are two major parameters in ATLAS that have a significant effect on the base current in the medium injection regime, these being the workfunction of the poly-emitter and the band-gap narrowing affect. These two affects are described in sections (a) and (b) below.

(a) Poly-emitter work function
If the poly-emitter is described as N.POLYSILICON in the CONTACT statement for an NPN device as described in section (iii) above, the work function of the poly-emitter is set to 4.17V and is correct for saturation doped n++poly-silicon. If, however, the poly-emitter is not saturation doped, the work-function will differ from this ideal and have a pronounced affect on the base current and current gain in the medium injection regime. The work function of the poly-gate can vary from 4.17V for n++ poly-silicon to (4.17V + Eg) for p++ poly-silicon depending on the position of the Fermi-Energy. Changing the work-function of the poly-emitter by just 0.1V from 4.17V to 4.27V can often halve the current gain in the medium injection regime as shown in the current gain plot of Figure 2. It is therefore very important to assign the correct value. The contact statement below assigns a workfunction of 4.27eV to the poly-emitter, whilst keeping the other parameters the same as before:



Figure 2. Effect of emitter contact workfunction on bipolar gain.


The poly-emitter workfunction can be calculated by measuring the position of the fermi-energy at the poly-silicon/silicon interface relative to the conduction band and adding this value to 4.17V. For example, if the Fermi-energy is measured as being 0.1eV from the conduction band edge, the workfunction of the poly-emitter set in the contact statement should be set to 4.17 + 0.1 = 4.27V.


(b) Band-gap Narrowing Effects
If the BIPOLAR parameter is stipulated in the MODELS statement in ATLAS, bandgap narrowing is included automatically. The inclusion of bandgap narrowing in the MODELS statement is strongly advised since this phenomenon has a significant effect on the current gain of the device. However, in order to validate the default Klaassen bandgap narrowing model, the Klaassen mobility model should also be used. This is activated by using the additional keyword KLA to the models statement. For example:


will correctly activate the Klaaseen bandgap narrowing model. The parameters in the Klaassen band gap narrowing model are user definable in the MATERIAL statement. There are three user definable parameters for the Klaassen band gap narrowing model. The BGN.E parameter has a linear dependency on doping concentration and has the default value of 6.92e-3 volts. BGN.C has a square root dependency with doping concentration and has the default value of 0.5. BGN.N is the value of doping where band gap narrowing effectively starts to take affect and has a default value of 1.3e17/cm3. The equivalent default setting would therefore be written as:

MATERIAL BGN.E=6.92e-3 BGN.C=0.5 BGN.N=1.3e17

These parameters can therefore be altered to modify the current gain of the device in the medium injection regime. For example, reducing the linear parameter from 6.93e-3 to 6.5e-3 is sufficient to cause a significant reduction in current gain in the medium injection region. The band gap narrowing parameters affect both collector and base currents but affect the base current to a greater magnitude. The most sensitive plot to see the effects of small changes to the band gap narrowing is a plot of current gain versus log of collector current. A reduction in band gap narrowing will result in an increase in current gain in the medium current injection region.

(v) The Base Current Profile - Low Injection
This is one case where there is an interdependency on one parameter, since the intrinsic base resistance not only affects the collector current in all regions(see above) but it also has an affect on the base current in the low injection region as shown in the gummel plot of Figure 3.

Figure 3. Effect of base doping profile n low injection base current in BJT.


For a small range of implant doses around the optimum, the base doping concentration will also affect the position of the knee and/or the rate of falloff of the base current in the low injection operating region of the device.This is most noticeable as a loss of current gain in the low injection region for the alternative standard plot of current gain versus collector current.An increase in the base implant reduces the intrinsic resistance and typically decreases the base current in the low injection region resulting in an increase in current gain for very low currents.

A similar affect to increasing the base doping is observed if the base doping is kept constant but the overall doping is reduced in the mono-crystalline silicon region of the emitter. The doping profile in the mono-crystalline region of the emitter can be tuned using three parameters in the process simulator.

The main physical affect of these Athena parameters is to change the doping profile of the emitter in the mono-crystalline silicon. These three parameters are:

1/ the grain structure in the poly-emitter.

2/ the dopant segregation effects in the poly-emitter.

3/ the dopant velocity across the silicon/poly-silicon boundary.

The first parameters will affect how quickly the dopant in an implanted poly-emitter reaches the silicon/poly-silicon boundary during the RTA diffusion and therefore affects the total diffusion of dopant into the single crystalline part of the emitter and hence the base width doping profile. The POLY.DIFF diffusion model is used.

The second parameter affects dopant pile-up at the poly-silicon/silicon boundary and hence the source doping concentration at the monocrystalline interface. Once again, therefore, this will affect the overall doping profile of the emitter in the mono-crystalline region of the device.

The third parameter affects the velocity of transport of dopant across the poly-silicon/silicon boundary with similar affects to the parameters above.

These parameters can be used to tailor the emitter doping profile in the mono-crystalline silicon region to match available measured data, usually in the form of SIMS or capacitance information. An accurate profile of dopant in the poly-silicon part of the emitter is not too important if measured data concerning interfacial dopant concentrations is available. This is because the work function of the poly-emitter is going to be set in ATLAS , the device simulator, by defining the poly-emitter as an electrode. All that is required to calculate the correct work function at the poly-silicon emitter is the interfacial doping concentration at the poly-silicon/silicon interface on the poly side of the junction. Setting the correct workfunction for the poly-emitter is described in section (iv-a) above.


Following the methodology presented here and by using a logical combination of tuning parameters available to the user in ATHENA and ATLAS, a good match for bipolar transistors should be obtainable for most device designs.

Since it is usually less problematic to match the collector current for all levels of applied base-emitter voltage compared to the matching of base current, the user will probably find that more time is spent tying to match the base current for very small and very large values of applied base-emitter voltage. The user should ensure, however, that great attention is placed on making sure that the correct process models are used in the process flow to reduce the overall uncertainty as to which parameters require calibration.