We have always had programmable analog circuits. Any circuit that makes outputs change as a function of inputs is in some way programmable. Examples include gain change as a function of voltage or current (automatic gain control, or AGC) and frequency as a function of voltage or current (a voltage-controlled oscillator). With appropriate signal conditioning, a clever designer can make parameters vary with time (an ultrasound time-varying gain "s" curve amplifier), or any other base parameter.
As we look at the spaces for analog functions that need to change parameters while in operation, the designs are becoming increasingly complex and unwieldy. Improvements in all electronic components that make them faster and more accurate while the devices operate at lower power levels per function mean that analog control functions also are becoming more precise. That precision comes at the cost of resources—more design time or more complex components.
Analog is always custom. Every application varies, which is why there are so many different analog parts out there. The measurements and controls in analog systems require many specialized components to fulfill the diverse requirements. Due to this demand for unique combinations of functions and parameters, many parts fill the analog world, each representing a wide variety of performance. The proliferation of products shows that one size or generic function doesn't fulfill the needs of all circuit designers. Instrumentation amplifiers, for in-stance, come in 20 to 30 variants that address all possible combinations of input and output parameters.
As the systems become more refined, the controls must become more responsive and flexible, but at a reasonable cost. Therefore, a designer may not want an external signal to control the function. In a data-acquisition system, you might want to get away from the full converter chain per signal, so all of the signals go into a multiplexer. The signals get adjusted in the common portions of the signal-processing chain. But an AGC may distort the critical parts of the signal if signal changes are of the same time scale as the AGC loop.
Enter Programmable: Depending on where the part is used in the system, programmable analog could mean different things. Instead of complex feedback loops and critical component values, designers want to migrate toward simpler architectures. Designers want to give the task to programmers because the crusty analog guru is getting crotchety and incorrigible, not to mention nearing retirement.
Mick Britchfield, product line manager for precision converters at Analog Devices, notes the trend of programmability in analog design. One issue is the need to create hardware and software that enable a high level of programmability without requiring the designer to become a DSP programmer.
The integration of more analog components into the systems requires more interfaces to functions like analog-to-digital converters (ADCs) and other mixed-signal functions. Vendors need to offer the right combination of analog functions to solve the problems in the analog spaces. This is both an integration issue and a technology issue. Silicon no longer limits the number of components and structure complexity.
Silicon solutions cover a wide range of programmable functions. Continuous functions can be controlled by an analog switch that changes some parameter in an on-off basis, or a circuit could contain a digital potentiometer that adjusts its operating performance.
At the other end of the range, the analog subsystems can include ADCs and digital-to-analog converters (DACs), plus some DSP to calculate the signal-processing algorithms. In between these extremes, the discrete programmable components and the data acquisition and conversion systems, are analog equivalents to programmable logic devices (PLDs)—arrays of programmable switches and some fixed functions that can be configured to make up medium- to high-complexity circuits on a single programmable chip (Fig. 1).
Due to the nature of analog signals, all systems need to work with fairly high dynamic ranges, and all of the system implementations still require continuous gain stages before much of the actual signal processing can take place. Analog designers face the challenge of finding the best mix of components and function blocks and putting them together in a stable, robust, and cost-effective system.
Tony Ochoa, product marketing manager for linear products at Xicor, states that the major trend in systems design is toward DSP functionality. Direct analog functionality is usually smaller and has lower power, but the designs are very complex and require a lot of math and physics. On the other hand, DSPs are simpler from a topology perspective. The DSP gives the designer mobility in the design portfolio and lets design sections be moved to other projects.
Many designers think that a one-chip solution is a desirable goal for all ICs, and that the upper bound for integration is a realistic limit for any set of functions. Higher-integration levels allow the user to benefit from smaller parts, greater complexity, and many other savings, including smaller board space, lower power, reduced tracking and matching issues, fewer external components, and less inventory.
John Pendergast, a product manager at Linear Technology, observes that in every analog system, designers need to consider the tolerances and drift characteristics of all external components. Designers must try to get as many parts as possible integrated into the analog chip to minimize tracking issues. The more highly integrated subsystems will tend to have all internal components track over operating voltages and temperature.
This system-on-a-chip (SoC) mentality is a growing driver in the design community, but the majority of SoC designs are "Big D/ little a" type designs, meaning that they're digital-logic dominated. Unfortunately, this logic-centric view means that the care and concern for analog performance is very reduced. Also, the digital designers think that all signal and processing requirements can be addressed by the DSP, forgetting that poor signal quality requires more work to clean up the signal before other processing can take place.
At the same time, the move in process technologies toward ever smaller features calls for lower supply voltages and more restricted operating and dynamic ranges for the components. The ability to process data at 24- or 32-bit precision may not make sense for a system that has millivolts of noise over microvolt signals. In addition, most analog functions don't use the smallest possible devices for the signal paths, but employ the device geometries from at least one or two process generations back.
The name of the game is balancing the tradeoffs. How much programmability versus complexity in design and cost make the most sense for a given application? If the signal processing is in the continuous time domain, only a little can be realistically integrated. If the signal processing is in the sampled data domain, the sky is the limit.
Full-Time Analog? Tradeoffs between continuous and sampled implementations make it harder to exclude either mode of operation. Continuous parts generally have faster responses—or less lag before response—but may have problems with matching and drift. In addition, for high-order filters, many components must be hand-selected to get circuit performance within specifications. The continuous versions of a filter or other analog functions use fewer components than sampled functions.
Sampled data representations in-clude switched-capacitor functions and DSP systems. Switched-capacitor functions offer very good tracking and matching characteristics, the fast responses of continuous functions, and the precision of digital systems. The switched-capacitor filters may need a band-limiting function before the switched-capacitor function. A smoothing filter might also be necessary on the output to return the signal to a continuous mode.
The completely sampled data systems require a band-pass or low-pass filter, track and hold or sample and hold, an ADC, memory, a DSP, and a DAC to bring the digitized signal back into a form that the real (analog) world needs for controls. These are a lot of components, with a fairly long acquisition and conversion cycle. But they offer high stability, repeatability, very high precision, and performance capabilities that can't be matched by continuous systems.
From Low- To High-Level Functions: Because many systems may need internal control of variable parameters, externally programmable components provide many capabilities not always available from the traditional alternatives. At the lowest level, programmable functions are available as discrete components that interface to the control system. Some include such functions as digital potentiometers and programmable gain amplifiers. These components can replace their mechanical equivalents and provide lower costs and higher reliability compared to the mechanical parts.
Tim Green, strategic development engineer for high-performance linear products at Texas Instruments, states that programmable analog functions have become an increasingly popular solution for improving the accuracy of real-world inputs without the additional cost and time associated with the traditional methods of potentiometers, select-in-test resistors, select-in-test capacitors, and laser trimming of module assemblies.
For instance, the potential savings in replacing a mechanical pot with a digital one are lower-cost components and lower-cost test and adjustment. The savings in test and adjustment costs will be significant because the board tester can set the pots as a part of board test, compared to the minutes for a technician or operator to set-measure-reset-remeasure a mechanical unit. The screwdriver for adjustment is replaced by a digitized equivalent—a control bar, bar graph, slider, or virtual knob.
Maher Matta, business manager for high-speed converters at Maxim Integrated Products, claims that with all the benefits of digitally processed programmable analog, designers need to get the signals into digital form as soon as possible. For example, RF systems can operate as full digital radios and demodulate any signal with just the appropriate software and algorithms.
The architecture is simplified because only one front-end subsystem is required to capture the RF and digitize it. This architecture eliminates multiple frequency conversions and filtering steps. The downside is that the software radio RF stages are very fast—2 to 5 GHz for the latest wireless local-area networks. Getting a converter and track and hold to operate at the giagahertz frequencies, with the bandwidths and tight jitter specs necessary, means the digitizing stage will use a lot of power.
At the data converter portion of the systems, the vendors are adding new functionality and capabilities to the components. Because the parts are mostly digital, increasing the integration doesn't severely affect the other areas of performance. In addition, vendors are integrating more converter functions into the base analog devices to make a whole analog subsystem.
Jim Todsen, data-acquisition product line manager at Texas Instruments, says that one trend in ADCs is integrating additional features into the same device. Although it solves many problems, integration alone may not be enough. On top of integration, programmability adds flexibility.
Programmability can mean several things. It can mean that the ADC can be "customized" to a specific application. Programmability can also mean the ability to use the same device as a single platform and to operate for different systems. Plus, it can mean that the device can adapt to changing conditions in the system.
A discussion of programmable analog technology would be remiss if it didn't include the latest entries into the field. Programmable analog arrays are moving into the design consciousness as an alternative to the difficult and temperamental collection of continuous components and the high-complexity DSP systems. Most of these arrays use merge memory and an analog switch matrix, configured as a crosspoint switch, to provide analog function blocks and reconfigurable in- terconnections. The volatile parts are SRAM-based, like FPGAs, and the nonvolatile parts are mostly based on electrically erasable (E2) technologies.
The latest products are the silicon and the EDA and software tools for creating and verifying the design. Analog arrays offer very high levels of programmability, from internal component parameter adjustments to overall configuration of the function blocks. Some have characterized these products as the analog equivalents of early PLDs, but the latest offerings are nearing the small FPGAs (Fig. 2).
Design flows require a new methodology and design philosophy to achieve the best results. Because the tools use higher-level system views of the analog functionality, designers need less experience in the cause and effect of circuit value and topology changes. The highly integrated components greatly reduce the need for critical matching, external component tracking, and tuning in manufacturing.
Stan Kopec, vice president of marketing at Lattice Semiconductor, claims that the programmable analog component market is still in its infancy, roughly equivalent to the state of programmable logic in the late 1970s or early 1980s. The market is still developing. Designers must change methodologies and design philosophies to make the market grow faster.
The first big necessary change is to bring a methodology like the programmable logic design tools to the analog design space. Designers require a suite of design-entry tools that enable the easy development of designs that take advantage of available silicon capabilities. Designers have to become familiar and comfortable with the idea of downloading a bitstream via a JTAG port to alter the circuitry.
Higher-level tools, instead of just Spice, let the designer program functions in silicon to make the final functions. The higher level of integration reduces component variability and makes the design much less sensitive to the parasitics and layout variables. The integrated components also improve tracking because almost everything is on the same substrate, minimizing temperature and voltage differentials. The integrated parts make the problem easier to solve and conquer.
An appropriate combination of tools and silicon can reduce the level of expertise needed to design analog circuits and allow designers with minimum knowledge of circuit characteristics and complexities to create working analog subsystems in a short time. Just like programmable logic parts, programmable analog arrays offer quick confirmation and verification of the design. All programmable arrays have associated design kits and evaluation boards to let the designer quickly check the analog system performance in a real circuit.
The overriding problem with generic analog arrays is that they may be a good idea but might not be economically feasible. The flexibility costs $2 to $8 per part. Also, designers can't have full-accuracy analog functions, which are application-specific, in a general-purpose analog array. Nevertheless, the latest programmable analog function blocks provide very high performance. Some of the parts even permit for on-the-fly reconfiguration. This ability to change circuit parameters and topologies as needed means that the designer no longer has to trade off specific functions and performance for cost or general applicability in the system.
|Need More Information?|
Analog Devices Inc.
Lattice Semiconductor Corp.
Maxim Integrated Products
National Semiconductor Corp.
Texas Instruments Inc.