Making good use of amplifiers to optimize analog IC extreme performance design

Microwave designers have used optimization techniques to improve and focus circuit performance in their designs for decades. Thanks to new technologies developed in the past decade, analog IC designers can now easily establish and efficiently perform optimizations on their designs.

Unlike previous circuit optimizers that had to be set up and run primarily in batch mode, these newer solutions are designed to make the setup and interactive use of the circuit design creation phase much easier. While many solutions only include one algorithm, some tools now offer many optimization algorithms and methods that can be applied specifically based on the specifics of the problem and the width of the design space. Many of these algorithms start from a user-defined starting point and search the design space for a local optimum. Other methods are able to search the entire design space for a global optimum.

Let's analyze an application example where the requirement for an analog IC is to amplify an intermediate broadband signal up to 2GHz. As an operational amplifier, this IC is always used in a closed loop structure, which is a real challenge at these frequencies. Therefore, the phase shift of the signal after passing through the amplifier must be kept to a minimum. Due to this high frequency requirement, the amplifier will be implemented using 60GHz silicon germanium technology.

The amplifier is designed to meet or exceed bandwidth and gain requirements while minimizing power and maintaining stability. Indeed, these requirements are very conflicting. Just meeting these specifications can take hours or even days of work for a designer, let alone finding the best solution. Often, in order to save time, designers have to settle for a barely acceptable solution that does not maximize the design's potential. This is where optimization can really shine.

In addition to bandwidth, other requirements such as gain, power, and stability must also be considered. In this case, power supply rejection ratio and preferred DC offset are also trade-offs in the optimization. Most of these goals are unequal constraints that must be less than or greater than a target value or line segment.

Once the measurement parameters have been defined, setting the optimization goal is easy. The user simply selects the measurement parameters that need to be measured during the optimization session and chooses whether to make it less than, greater than, or equal to a certain value (or a range of values ​​over a certain frequency or time range, if applicable).

Once these objectives, weights, design parameters, and constraints are defined, the optimizer is ready to run. Since this amplifier has discrete design parameters, either pointer or randomized algorithms can be applied. Pointer algorithms are more appropriate in this case because they are generally more effective for nonlinear problems where simulation run time is expensive. Analyzing the results after the optimizer runs 50 iterations will show that the cost function has been substantially improved. After making the final adjustments, the optimizer runs a total of 100 iterations in about 30 minutes to further optimize those parameters.

At this point, the objective weights can be refined to improve performance at the expense of some other requirements. Also, as the optimization process continues, some design parameters become less important and then no longer useful. The optimization process is continued with another 100 iterations/iteration refinements, ultimately resulting in an overall trade-off. This interaction is critical to maximize optimization. The optimization process takes several hours, but the designer is confident that the full range of trade-offs can be achieved to maximize the performance of the amplifier.