It's been about 40 years since the first monolithic op amps were introduced. Those first devices, some of which are still in production, were designed exclusively in bipolar technology. Today most single-chip op amps are still designed using bipolar transistors or their close cousin, the junction gate field-effect transistor (JFET), even though most 40-year-old IC innovations have long since become CMOS turncoats. Perhaps even more astonishing, no company has yet invented a one-size-fits-all device. In fact, there aren't five-, or ten-, or even one-hundred-fits-all devices. Try over 1,000! Texas Instruments alone lists nearly 700 op amp products in its web site search engine. What is it that keeps 21st century op amp designers living in the '60s? Where is the Pentium 4 of op amps? The answers lie deep inside the thousands of manufacturer data sheets: The Specifications.
Specifications: General versus Application
Although not found in every data sheet, cumulatively there are at least 30 different specifications that relate to the performance of an op amp. Each specification falls into one of two possible categories: General and Application Specific. The General category, shown in Table 1, is composed of items like input offset voltage and current, common mode rejection ratio (CMRR), and others. The Application category, also shown in Table 1, is composed of items such as noise density, gain-bandwidth product, output load drive, supply voltage, and current. What sets these two categories apart is the definition of what constitutes the best performance.
All of the specifications in the General category have the same definition of “best.” The best offset voltage for all applications is that one closest to zero volts. The best CMRR for all applications is that one closest to zero volts/volt. General specifications primarily relate the degree of mismatch between circuit elements on a properly designed IC. If the input differential pair of an op amp is perfectly matched, the input offset voltage, offset current, and drift are nearly zero: a universal best. There are also many op amp products available where a post-manufacturing trim is performed to remove the random mismatches that result from normal processing. The result is a dramatic improvement in General specification performance for some parameters.
Conversely, the Application category has no single definition of best. There are only tradeoffs based mostly on electrical theory. For example, the best broadband noise performance is generally available on parts that exhibit relatively high operating currents. If your application demands very low operating current with no concern for broadband noise, your definition of best may have you choosing a part with 150nV/√Hz of voltage noise density at 5μA of supply current. On the other hand, your colleague down the hall may need 1nV/√Hz but has a substantially higher operating current budget. She has a substantially different definition of best. Given this very fluid definition of best, the Application category has become the Achilles heel of the op amp industry and the reason why there are more than one thousand different amplifiers to choose from when specifying a device.
A good method for selecting an op amp from myriad search-engine parameters is to start by defining the minimum acceptable General specifications for the circuit application. During this step of the selection process it's extremely important to not over-specify the performance limits. There are two reasons to be cautious.
First, the highest performance General specifications are typically available on devices that have circuit topologies optimized primarily for matching. An example of such a topology is the OP177 manufactured by Analog Devices and shown in Figure 1. While the OP177 has impressive matching performance in the form of low offset voltage and high CMRR, such an input topology is not preferred for high speed or large common mode input range.
On the other hand, the LT6220 manufactured by Linear Technology and shown in Figure 2 has an input stage capable of handling signals over the entire power-supply range and at frequencies 100 times higher than the OP177. Alas, that same LT6220 exhibits a CMRR that is a factor of 100 less than the OP177. The point here is that by over-specifying the General requirements, you will be left selecting from a group of parts that severely restrict the potential performance of other aspects of the circuit.
Secondly, and with only a few exceptions, those devices with the highest General specification performance have the lowest manufacturing yields and/or require post-manufacturing trimming. Both of these translate to a higher manufacturing cost and a higher sale price. One exception, although not a classical op amp topology, would be a chopper-stabilized or auto-zeroed amplifier where sampled-data techniques are used to eliminate mismatch yielding issues.
Once a search list of amplifiers with acceptable General performance limits is established, the list can be culled down to a best-fit part based upon the Application performance requirements and, ultimately, cost.
CMOS is King. NOT!
For the past 20 years analog circuit designers have been slowly moving numerous analog IC product lines to CMOS technology and in some cases leaving bipolar as a footnote in history. One can see this transition in areas such as analog-to-digital and digital-to-analog converters, low-frequency filters, the complete receive and transmit channels of single-chip radios, portable power management, sample/track and hold amplifiers, and others. Conspicuously absent from this list, though, are classical high-performance op amps; arguably the most analog of all analog circuits.
Sure, sprinkled among the endless lists of op amps you can find a multiplicity of modern, nonclassical, auto-zero style CMOS amplifiers with magnificently high common-mode rejection, unheard-of low offset drift, and virtually no low-frequency noise, more commonly referred to as 1/f noise, but outside of very low-frequency applications you'll be sorely disappointed by the switching noise, aliasing, distortion, overload recovery time, and other unexpected and annoying artifacts that accompany those specialized architectures. Since the recent turn of the century, CMOS amplifier manufacturers such as Texas Instruments, Analog Devices, Linear Technology, and others have made significant inroads supplanting bipolar as the leading technology for high-performance op amps. Despite their impressive results to date, they must ultimately overcome three major hurdles before they will be successful.
Perhaps the most often cited hurdle is the superior elemental matching obtainable with a bipolar process. Poor elemental matching in a process restricts the General specification performance of an analog IC. Although it's beyond the scope of this article to offer a detailed technical presentation to explain why this is true, consider that PSPICE, a popular circuit simulation software program, requires over seven pages of mathematical variables to describe the operation of a MOS transistor to a computer. Compare this to the bipolar model used by the same program where about one page of variables is adequate. Needless to say, if there are a substantially greater number of variables that define the behavior of a MOS transistor above that necessary to define a bipolar transistor, the relative overall performance variation is bound to be greater.
Equally as limiting as poor matching is the lower transconductance efficiency of MOS transistors. Low transconductance manifests itself in an op amp as high transistor broadband noise, low output load driving capability, and low current gain. Here the difference lies in how the transconductance of a transistor scales with bias current. In a bipolar transistor the scaling is linear, while a MOS device scales with a much weaker square root dependence on bias current.
Finally, MOS transistors are notorious for being a source of high 1/f noise. It's not unusual to find CMOS op amps with 1/f noise corner frequencies approaching 100kHz while a similarly biased bipolar op amp operates with a noise corner near 1Hz. Here the problem lies with how MOS transistors are manufactured. Whereas a bipolar transistor exhibits gain through a conduction path buried well below the surface of the IC, as shown in Figure 3, a MOS transistor has its gain occur at the surface and, most importantly, in a region with numerous noise-generating defects, demonstrated in Figure 4. Perhaps less understood than the simple 1/f noise effects, these defects contribute additional performance limiting artifacts such as input offset voltage-recovery time following a large voltage step transient.
The times, they are a-changing
To borrow a line from Bob Dylan, the times are a-changing. In the past few years, op amp manufacturers have begun introducing several new high-performance CMOS op amps. Chief among these are amplifiers designed for high speed with gain-bandwidth products that are knocking on the door of 100MHz. Despite these admirable new developments, the scene for highest performance op amps has not changed much in regards to the three hurdles described earlier. Bipolar still leads the pack and is holding on strongly, although perhaps not for long.
In late 2004, Texas Instruments introduced the OPA727 MOS op amp, a 20MHz bandwidth amplifier with a bipolar-class input offset voltage and, perhaps more importantly, an offset voltage drift of 1.5μV/degree-C max. To obtain this result TI trims the offset voltage at different operating temperatures after the device is packaged. In true CMOS style, though, the low-frequency noise for the OPA727 is 10μVp-p.
This past September Linear Technology introduced the LTC6078, a CMOS-input amplifier with superior offset voltage and drift characteristics. While not strictly an all-CMOS amplifier such as those sold by Texas Instruments and Analog Devices, this device employs CMOS transistors as the input differential pair for the same reasons: to obtain very low input bias currents, and to maximize the input common-mode voltage range.
To combat the inherent CMOS problems of poor matching, high 1/f noise, and low transconductance, Linear Technology utilizes extremely large devices in the critical areas of the LTC6078 signal path. As a comparison, consider that the device's common-mode input capacitance is nearly 20pF whereas a comparable low-noise, low-offset bipolar op amp exhibits less than 5pF. Although, even in this area, Linear Technology has introduced technology in other CMOS input-amplifier products, such as the LTC6241, that partially cancels the capacitive effects associated with such large devices.
Also in September, National Semiconductor introduced several new precision CMOS input amplifiers built with its new, proprietary, VIP-50 process technology. Here the company is leveraging advanced process technology in addition to circuit design prowess to differentiate its CMOS amplifier product line.
Despite the use of large MOS devices and capacitance reduction technologies the overall performance of CMOS precision amplifiers is still less than that obtainable from a classical precision bipolar amplifier.
Bipolar not going away soon
While CMOS technology may be the cheapest IC manufacturing technology, it takes the will of a company and market pressure to move product research and development in that direction. The experts in analog op amp design reside in companies whose product lines stem from a bipolar family tree dating back decades. These veteran op amp designers take huge pride in the power of a single bipolar transistor that admittedly can do a better job of amplification than a 10-transistor CMOS circuit. The truth be told, this is probably more a statement about designing CMOS op amp architectures that try to mimic those that work best for bipolar rather than inventing new architectures that trade transistor count for equivalently sized area. To ultimately topple bipolar op amp technology from its perch at the top of the op amp market will undoubtedly require new architectural approaches rather than simply repackaging past bipolar success stories. Texas Instruments' e-Trim, Analog Devices' DigiTrim, and Linear Technology's capacitance-reduction technologies are certainly steps in that direction.
In the end, the cold hard facts are that if you can design an adequate analog product entirely in CMOS, that's where it will ultimately be. The same cannot be said about bipolar. Said another way, bipolar is a technology of last resort. Today in op amp design, all the leading analog IC manufacturers still avail themselves of that last resort quite frequently and in more than 1,000 different ways.
J. Scott Elder of Analogue Integrated Technology is a analog IC design consultant based in Orlando, FL. He can be reached at .