Something a bit more current;
Appendix A. A Brief Discussion Of Competing RFIC Technologies
1. Introduction
The fundamental foundation of this technology development is the belief that a low complexity, low cost, high performance, Silicon, pure Bipolar process can compete on the basis of the factors cited in the body of this report. The competing technologies are Si BICMOS, Si RFCMOS and GaAs Bipolar or HEMT, the relative merits of these is discussed briefly here.
2. Silicon or GaAs?
This is the subject of a large tome not a short paragraph, but some relatively uncontroversial points can be made. There is little doubt that the best raw RF performance can be obtained from a GaAs High Electron Mobility MOS transistor (HEMT) and these devices can be combined to form small RF MIMICs, but, these devices will always be expensive and levels of integration will always be very low.
The advent of SiGe technology means that, in principle, there is little difference in potential RF performance between Si Bipolar and GaAs Bipolar devices. Future development may also see the development of SiGe HEMTs. A relevant point is this discussion is the relatively poor thermal conductivity of GaAs compared to Si (0.46 to 1.5 W/cm- C). In many applications it is critical to sink heat away from active devices and often Si wafers are thinned to between 100 and 200 microns thickness to facilitate thermal conduction. The poor thermal conductivity of GaAs makes this much less effective and the fragility of GaAs generates many problems in the handling of thinned wafers.
Another real difference between the two materials arises in their relative ability to incorporate support functions onto the RFIC. Considering a typical PA, functions like power level stability control and power level programming can be integrated relatively easily, and at low cost, in SiGe but not in the GaAs case.
On the receive side, the ability of SiGe to make complex phase cancelling active mixer circuits with positive gain eases the noise and gain requirements on the LNA. It also reduces the Local Oscillator power level by orders of magnitude with a consequent large reduction in LO breakthrough problems.
3. BICMOS In Context
Considerable effort has been devoted world-wide since about 1990 to the conjoining of CMOS and Bipolar processes in order to mix the good high frequency, high current drive capability of Bipolar devices with the dense logic of CMOS. In this way it was thought that complex functions could be manufactured and the goal of “System On A Chip” achieved. Many such processes exist today but it is fair to say that the market penetration of BICMOS has not been as extensive as was hoped. Some of the factors which have affected this are discussed below.
A key element is the cost overhead related to the addition of the Bipolar steps to a CMOS process. Down to 0.5 technology there were many common process stages and so the extra cost was quite limited. However, both IBM and STMicroelectronics have published papers recently advocating the separation of the CMOS and Bipolar process steps, this will produce the best device performance but represents a formidable increase in costs.
A second concern for BICMOS is time to market. Inherent in the SOC approach is a major design task, not just in the initial work but in debugging and updating. In many markets, for example cellular telephone, the product lifetimes are less than two years, this places a difficult burden on a product design team.
A final element in this tale of woe, from an RF perspective, relates to the RF optimisation of the Bipolar device. This is a difficult task at any time, if some degrees of freedom are eliminated by the CMOS starting point, then it is almost axiomatic that the all-round performance of the BICMOS Bipolars will not be as good as that obtained from much simpler, pure Bipolar process. This may seem to be a very negative analysis, but, one characteristic of the microelectronics industry is that almost all technologies find their application and this will be true of BICMOS
4. RFCMOS
One of the clear trends in the recent past has been to design RF functions in pure CMOS technology. Indeed there are companies designing products for 5GHz WLAN applications. Considering device fundamentals, the best MOS device will not have the same transconductance (gm) per area as a Bipolar transistor. Similarly, the RF performance Ft and Fmax will not be as good. However, it has been demonstrated many times that good circuit design can overcome
technology deficiencies.
If we compare the dynamic performance of MOS and Bipolar transistors it has been shown that parallelling MOS devices for more power capability does reduce the power gain whereas this does not occur for Bipolars. Consequently MOS performance will be poorer where high dynamic range or significant power is required.
To date RFCMOS has been successful at relatively undemanding applications such as pagers and little real information is available on the kind of circuit needed for PERLA. Consequently, judgement on RFCMOS should be suspended for the present.
5. Conclusions
Essentially there are no conclusions, the technology choice for a product is a result of cost, opportunity, design capability and availability. Consequently the same product may well be made in a variety of technologies by different companies.
What can be said is that the PERLA project choice of an optimised Bipolar technology for critical WLAN functions and then adding further capability via multi-chip or System In A Package is one of the potentially successful product routes.
perla.intranet.gr |
