神马文学网Tolerances
来源：百度文库 编辑：神马文学网 时间：2024/04/26 05:01:16
Simulating inaccuracies and fluctuations in environmental conditions
- Temperature
- Process fluctuations
- Operating voltage
Temperature
Already the name tells that SPICE (Simulation Program with Integrated Circuit Emphasis) has been designed especially for designing integrated circuits. Without simulation tools designing integrated circuits wouldn’t be economically possible. Because of the time of design, SPICE2G6 has also assumed that the integration is only done for silicon. Because silicon is a very good conductor (unlike GaAs), the basic assumption is that the entire circuit has the same temperature. The temperature can be controlled with the .TEMP-command (in HSPICE also with the SWEEP-command on the analysis line, e.g..TRAN 10n 1u SWEEP TEMP -40 110 10). The temperature that the simulator has set, affects the models of both passive and semiconductor components through different temperature coefficients. In addition, at least in Hspice, a DTEMP-parameter has been added to component models. With the help of this parameter the temperatures of the components can differ from each other.
When power levels increase, the components start to warm themselves. However, this phenomenon is very dynamic: the warming-up and the cooling of a transistor happens with a delay. This of course distorts the signal. SPICE2G6 doesn’t take account for this heating effect, but in APLAC, for example, it can be defined for all the components. In addition you can make an electro thermal simulation with APLAC, which pays attention to both electric and thermal connection by adding a thermal circuit to the simulation. However, similar functional models constructed from a sub-circuit can easily be added to Spice-based simulators.
Because of the dynamic sensitivity of the heating phenomenon, you’ll have to measure dynamic characteristic curves as pulse measuring besides the normal static characteristic curves for the power semiconductor components (and for all GaAs-components). This has to be done so that the effects of the heating can be modeled correctly. Unfortunately most of the GaAs MESFET models don’t take account for the heating effects.
Process fluctuations
The process companies usually give several transistor models for designing integrated circuits: typical (nominal) models and a varying number of different worst case –models. The worst cases are usually at least in terms of power (worst case power) and speed (worst case speed). Respectively there often are parameters for the corner parameters of the logic level (worst case one and worst case zero) for simulating CMOS-logic circuits. In that case the NMOS and PMOS parameters differ as much as possible from each other. The worst-case models come as a by-product of the process control: if the parameters extracted from the processing round don’t fit the gap of the worst-case models, the process round is discarded.The process fluctuations can be represented to the simulator also as statistical distribution, which can be used in the so-called Monte Carlo –run. The Monte Carlo –run, as you can guess from the name, simulates the circuit several times by randomizing new values for each parameter that has its statistical distribution defined. This way you can get a more reliable picture of the fluctuation of the circuit’s performance regarding the parameter fluctuation of the components. The situation can be that the minimum and maximum values of the circuit’s performance parameters are gained with some other combination of the process parameters than the worst-case corner parameters.
Modeling the parameter divergence of the Monte Carlo –analysis realistically is another problem. When designing integrated circuits there are in practice correlations between the parameters and there might not even be equivalency for the parameters of the semiconductor components. For example in CMOS-processes many of the process steps are common for both NMOS and PMOS transistors. In BiCMOS processes there can also be common process steps, and thus correlation between parameters, for bipolar and MOS transistors.
There isn’t Monte Carlo –analysis in SPICE2G6. The only way to do variation simulations is to vary the components or the model files with the .ALTER-command. In all the commercial Spice-type simulators Monte Carlo –analysis can be found either from the simulator or the user interface. For example the old Accusim ran SPICE2G6 and randomized new netlists for it and collected the results graphically as the result of Monte Carlo –analysis. Similar text-based Monte Carlo interface is the SWITCAP for SC-simulator that is made in the Electronic Circuit Design Laboratory.
The fluctuation and disturbance of the operating voltage
The operating voltage of circuits fluctuates always according to load, warm-up, component tolerances of the regulator and other disturbances. For this reason a range of fluctuation and the power supply rejection ratio, PSRR, must be specified for circuits. The range of fluctuation has to naturally be taken account for, as well as the temperature range. Thus there needs to be a margin of safety of about 100-200 mV, because the inaccuracies of the semiconductor models are at their highest at the boundaries of the area of operation.PSRR can be analyzed more easily with small-signal analysis by connecting the small-signal impulse to one of the operating voltages at a time. However, when simulating very symmetric connections you’ll get too optimistic simulation results, because the signal can be totally cancelled out in two identical signal paths. This of course is possible only in ideal simulation. You can hope to get results of the more realistic order of magnitude by, for example, “guessing” a suitable input offset voltage of the amplifier. The same goes of course also for simulating the common-mode rejection ratio, CMRR.
Updated 18.08.2005.