Analysis tools can’t design analog circuits
By Ron Mancini -- EDN, February 18, 1999
The last few years have seen a strong trend toward computer-aided analysis. Readily available, cheap computers; packaged Spice-based programs; semiconductor Spice models; and university training encourage new engineering graduates and beginning circuit designers ("newbies") to immediately jump into the uncharted waters of computer-aided analysis. Unfortunately, the results of this trend are more lost souls than working analog circuits. I support the semiconductor Spice-model effort because it fills my rice bowl, but I am sorrowful about the lost souls. About 30 years ago, during initial discussions about computers, somebody explained the mnemonic "GIGO," which means garbage in, garbage out. The newbies have a GIGO problem. They can't wait to start the computer analysis; thus, they grab almost any circuit configuration and begin analysis. Often, the starting circuit configuration can't perform the function they need, but they spend weeks analyzing this configuration in minute detail. Then, they review piles of data in an attempt to justify their efforts. This process is wasteful because they started with garbage in, so, by definition, they can get only garbage out. This scenario is the lost-soul trap; the volumes of data impress the newbies, so they spend weeks proving nothing to themselves. Now, if the newbies would analyze an approximate model of the circuit, they would quickly discover the circuit's inadequacies and make circuit-configuration changes. Simple models are easy to plot in the "s" plane, so you can quickly evaluate stability, overshoot, and performance. You can discover circuit-configuration inadequacies before analysis; thus, the changes to the circuit configuration, the model of the circuit, and the plot of the results eventually create an adequate circuit configuration for the job. The masters of analog-circuit design say that creating the circuit configuration is the artistic portion of circuit design. Therefore, it makes sense to consume the most time and brainpower creating the circuit configuration before any detailed analysis. Computer-aided-analysis programs don't encourage creative thinking about circuit configurations; rather, such programs provoke newbies into prematurely jumping into analysis. To avoid bunches of e-mail telling me that this situation doesn't happenat least not to experienced circuit designerslet me tell a story. I attended a meeting whose objective was to find out why a current-feedback op amp was oscillating. The designer presented reams of computer-aided-analysis data, all of which concluded that the op amp couldn't oscillate. The application engineer presented lab data proving that the op amp was oscillating under all operating conditions. Impartial parties like me were caught in the middle. I asked to see an s-plane transfer function of the op amp, and the designer informed me that he didn't have one because the circuit was too complicated to calculate a transfer function. I was shocked! Here was a person who was analyzing a circuit configuration containing dozens of transistors, and he didn't have an expectation of their performance. An outstanding case of GIGO! Another problem with computer-aided analysis is that it works only in a dreamland of ideal conditions. Computer simulations of high-frequency ICs can barely predict the circuit performance because of parasitic components, lead parasitics, wiring problems, and simulation problems. When a reduced-component Spice model represents the IC, the guessing game really begins. Some Spice models bear little resemblance to the IC (Reference 1). When you factor in this difference with the lack of preparation before analysis, this scheme guarantees failure. Computer-aided analysis is a wonderful tool, and I would not want to be without it. But, it is nothing more than a powerful calculator. When the calculator replaced the slide rule, you didn't expect the calculator to do your thinking, so don't expect computer-aided-analysis software to do your thinking, either. Also, test is mandatory; before releasing a design, test for the design parameters and user parameters. Reference
|
|||
 Ron Mancini is an applications engineer at Texas Instruments. You can reach him at 1-352-568-1040, e-mail rmancini@ti.com. |
|||
Talkback
