Archives | Feedback

Germanium Doping Enhances Silicon Performance in Wireless Communications
by Nick Smears, STMicroelectronics

Introduction

Silicon-Germanium (SiGe) has permitted silicon technologies to eat further into markets that were previously the stronghold of gallium arsenide by improving overall performances at higher operating frequencies. It has also improved circuit performances in areas that were already dominated by silicon Bipolar-CMOS (BiCMOS) such as mobile communications. These improvements are permitting systems to achieve cost reductions in these highly cost- sensitive applications.

Many publications on the properties of silicon-germanium heterojunction technologies have either concentrated on actual results obtained from certain implementations. This article attempts to explain the physical effects simply and show how they translate into concrete benefits at a system level.

Most of the discussions about silicon-germanium technologies tell us about the performances that have been achieved without going into the 'how'. Indeed, we are sometimes left wondering what the actual advantages are compared with pre-existing competing technologies. This article concerns itself solely with the place of SiGe in wireless communications. It is true that SiGe is finding an important place in the world of data transmission, but this is outside the scope of this article.

As pure CMOS radios are restricted so far to research papers and certain specific applications such as Bluetooth, it is fair to say that the competing technologies are classical bipolar junction transistors (BJT) (nearly always in BiCMOS) and gallium arsenide (or related III-V compounds).

First, we should examine where technologies using SiGe are being positioned relative to classical bipolar and gallium arsenide. Figure 1 shows the placement of various 'wireless' communications standards and an approximate estimation of the market share of the three dominant RF technologies today in the wireless market.

Figure 1: Technologies and Wireless spectrum allocations
Click for larger image (48KB)

Cost Optimization

Mobile communications devices, which are often considered as consumer electronics, are subject to such price pressures that cost is inevitably the main driver. Specifications have to be met, but performance improvements are always translated into cost savings.

Cost optimization has to be made by looking at the whole system, trading performances in one area against others elsewhere and, above all, making localized gains evident at system level. This may even mean that choices are made in one subsystem to allow benefits to be reaped in another.

For example, an improvement in the noise figure of the lower frequency end of a receiver chain could mean that the gain of the higher frequency end could be reduced, thus saving power. The lowering of the signal levels could also permit a relaxation in the distortion specifications of certain elements, possibly resulting in power savings or decreasing component costs. Additionally the lower noise and lower signal levels could lead to relaxing of filter specifications making these cheaper.

Power consumption has a direct and significant impact on cost. These are battery-powered devices with highly sophisticated energy management systems; reducing power consumption in the radio can be used to minimize battery or energy management costs. Batteries, it should be noted, represent a significant part of the overall product cost. On the other hand, increased standby/talk time could mean increased sales and therefore economies of scale.

Previously, the radio of a mobile communications device contained several gallium arsenide ICs at relatively low integration levels and with a much lower proportion of silicon. However, in recent years, silicon bipolar junction transistors have improved enough to allow them to push GaAs out from all but two functions - the power amplifier and the RF switches. Figures 2 and 3 show the evolution in the integration of GSM radios. This has been driven by the cost advantages enabled by silicon BiCMOS in terms of integration possibilities and its relative cheapness compared to III-V technologies.

It is interesting to note that GaAs still has the edge in terms of speed and noise figure - this underlines the strength of the argument for overall system cost.

Figure 2: Situation in 1995
Click for larger image (38 KB)

Figure 3: Situation since 2000
Click for larger image (35 KB)

Advantages of SiGe

So what are the advantages a silicon-germanium based technology over a classical (BJT) one of equivalent geometry? In short, it provides a big improvement in one parameter, the forward current gain, β. This increase can be sacrificed for two other improvements. To understand this slightly opaque statement, one has to remember that a bipolar transistor is a series of compromises and trade-offs. Modifications that improve the frequency performance either reduce the gain or worsen the noise performance. What improves the noise performance conversely tends to degrade the speed performance or the gain (or both); this is explained in detail in the inset.

The frequency response is characterized by the cut-off frequency (f_T), which is the frequency where the β drops to unity (Figure 4) and the f_MAX, which is the same thing, but for power gain.

Figure 4

The addition of germanium (for more information see: "How Germanium actually makes your transistor better") to the base of a BJT, turning it into a heterojunction bipolar transistor (HBT), permits a very large gain increase. This extra gain can then be traded against improvements in frequency response and noise factor. The base doping level can be increased to improve the noise factor, which also means a narrow base can be used to give a high f_T. Because the base is more highly doped, the fMAX is improved.

Table 1 shows a rapid comparison of typical performances for current generation processes that are in production. For III-V, field effect technologies such as HEMTs are also included.

Table 1: Summary comparison of SiGe, Si BJT and III-V technologies

[1] Wireless oriented technologies. Higher f_T's have been announced but on technologies that sit better with data transmission applications for cost reasons.

[2] PHEMTS; III-V HBT's have similar, if slightly better NF's to SiGe BJT's

Note 2 leads us to an interesting point: you can't have everything all at once. Technologies rarely provide all of their specified maximum performances at one time. An example of this is the RF small-signal transistor market, where it is common practice to quote the gain at one value of collector current, the noise figure at another and the IIP3 at (sometimes) yet another. An effect of introducing SiGe into STMicroelectronics' 0.35um BiCMOS was to reduce, to some extent, this phenomenon, i.e. the performances were achievable under one set of bias conditions. The STw0605D LNA demonstrator board is available.

The improvements made in BJTs have essentially concerned transistor geometries and the introduction of the poly-silicon emitter a few years ago. But it is becoming more difficult to achieve further significant improvements in both speed and noise for silicon BJT technology. Any improvements are made with increasing cost penalties, such as very fine geometries and complicated structures requiring more photolithography operations or trickier processing.

Solutions were sought to obtain significant improvements in silicon technology (to keep the overwhelming advantages of integration) without increasing the cost too much. Generally, SiGe BiCMOS technologies are settling into two camps: those aligned to mobile communications, and the more expensive ones with higher performances more suitable for fiber optic and other data transmission applications. Those in the second camp offer f_T's of up to 100 and CMOS geometries of 0.25um and below. Mobile communications RF applications rarely require more than about 40GHz f_T's or very high logic densities, but do require that any additional cost is small, when compared to existing technologies.

Impact for circuit performance

The main benefits, compared to Si BJTs, are seen in the areas of power saving and noise performance.

Power savings, passing from Si to SiGe, can occur in a number of ways. In certain situations, this is simply a result of the f_T/f_MAX increase. Figure 5 shows typical curves of f_T with collector current for BJT and HBT transistors. As can be easily inferred, a higher f_T means that the transistor can be biased with a lower IC for a given gain at a given frequency. Another way of looking at this is to compare the operation frequency, now around 2GHz, to the f_T, which is 20-30GHz for Si BJT. This puts us on the slope region of the β vs. frequency curve (Figure 4) and so the AC gain is lower than the static β; a 45GHz f_T is enough to put us back on the plateau.

All this means that somewhere in the chain, less power is needed for a given signal level.

Figure 5

In certain situations, power saving is more closely aligned to the higher transconductance of SiGe, compared to pure Si technologies. To meet circuit specifications, such as linearity for something like a mixer, a certain transconductance is needed. Transconductance increases (up to a point) with collector current. Therefore, quite simply, one can meet specifications with lower power levels.

Significant improvements in noise performance have been achieved with the introduction of SiGe. In moving from BJT to SiGe HBT's minimum transistor noise figures have been reduced as much as 0.2dB. But it is important to note that even noise is not so simple. Broadly speaking, there are two aspects - low frequency (1/f) noise and broadband noise. Although III-V HEMT technologies win on broadband, silicon wins on 1/f noise. Low frequency noise is important from a system point of view because it can fold up to RF and be seen as phase noise.

Thus SiGe provides an attractive compromise with good 1/f performance and an improved broadband performance as a result of the reduced R_B. What is most significant is that good circuit noise figures are achievable along with satisfactory performances in the other specifications.

Improvements in linearity have been seen when moving from Si BJT technologies to SiGe. This is a very complex issue because of the large number of factors that come into play. Many of these are related to properties of the implementation, so comparison with fundamental physics is difficult. It has been suggested that the improvements seen are linked to changes in the behavior of the base-collector junction.

Table 2 shows the characteristics of ST's 0.35um SiGe BiCMOS. Two ranges of transistor are provided, 3V and 5V. This is an interesting example of the effect of how doping levels and profiles can be adjusted to obtain the parameters desired.

Table 2: STMicroelectronics' 0.35um SiGe BiCMOS

Analog Main | Product of the Week | Columns | Editorial | Tech Notes

		Click here to get your listing up.