In talking to people in the electronics business, I find a general agreement that competent analog designers are hard to find. Some think it’s the fault of the engineering schools. One CEO at an analog chip startup complained that most of the candidates he interviewed were little more than “glorified programmers.”
My own view is less extreme. Most of the analog jobs these days are inside chip companies, many as applications engineers, needed inside the organization to make sure that reference designs for new IC products are bulletproof, or externally, to train customers, mostly in Asia, and sell them on the advantages of new products.
The analog engineers that the CEO is looking for tend to be half-engineer/half-physicists with multiple or advanced degrees and several summers of responsible hands-on experience in co-op programs at established companies that tend to hire them as soon as they become available, making them inaccessible to people like that startup CEO.
This report is about two aspects of that situation. One is board-level circuit design, using traditional analog components but with ever advancing performance specs. Analog IC companies provide a wealth of design tools. We’ll also review the downloadable and in-the-cloud tools from Analog Devices, Intersil, Linear Technology, Maxim Integrated Products, and Texas Instruments, including its recent National Semiconductor acquisition.
The other aspect is analog IC chip design at the cutting edge of performance. While the tools for board-level design tend to focus on applying mathematical analysis to bench-bench proven layouts, design at companies such as Linear is much more a matter of masters and apprentices, working in an intensely collaborative environment where observing the physical behavior of the most fundamental circuit elements delivers the greatest rewards.
The Grand Giveaway
The board-level design tools that semi companies make available are wonderful. In general, users know that the fundamental layouts have been lab-analyzed at all possible design corners. The layouts, when available, are designed to minimize crosstalk and noise, and the actual circuit behavior matches the Bode plots.
The GUIs can be extremely helpful in accelerating design optimization, the bills of material (BOMs) can be complete, down to actual disti pricing, and a key-click today can bring a complete design kit in tomorrow morning’s FedEx delivery.
In all cases, you’re going to be asked to register. That may seem off-putting, but despite all the things I’ve registered for, I’ve never been pestered by sales calls.
ADI’s Circuits from the Lab
Analog Devices’ Circuits From the Lab comprises a number of reference circuits that the company’s applications engineers have studied intensively in the lab to guarantee that they are essentially bulletproof and suitable for whatever application users have in mind.
When the application opens up online, users are presented with four tabs: one for circuit type, another for specific optimization criteria, and yet another for the specific application. The last tab accommodates any additional filter criteria that users might want to add.
As each tab is selected, appropriate screens pop up with check-boxes. For a trial run, I began by clicking the boxes for “General ADC drivers,” “Single ended,” and identified the application as a “Sensor Interface.”
The next set of choices was for specific optimization criteria. I clicked the boxes for “High-Resolution” and “Low Noise.” Then I got to dig deeper into the application. I rejected “Aerospace,” “Automotive,” “Medical,” and “Consumer” and picked “Building Technology,” specifically, “Programmable Logic Controllers or Distributed Control Systems.” This led to a selection of 22 circuit designs.
Essentially, I got 22 application notes, or documents in the form of application notes. The one I selected for further study bore the title “Flexible PLC/DCS Analog Output Module Using Only Two Analog Components.”
Click once and that document appears. It contains a description, circuit diagrams, and BOM information, along with graphic test data for the circuit and explanations of any cautions or special considerations the Analog Devices apps engineers think you should apply in laying out the circuit.
Even though Texas Instruments and National Semiconductor are now a single entity, the analog design tools that those companies created—WEBENCH and eLab—are both available on the Web (see “WEBENCH And eLab Online”). Generally, TI is treading gently as it integrates its new acquisition (see “Nine Things You Probably Want To Know About TI’s Acquisition Of National Semi” at www.electronicdesign.com).
For WEBENCH, I decided to design an active filter. Users begin by selecting the filter type, then defining the transfer function. I decided to make a band-stop filter (Fig. 1a). The first step was to specify attenuation in terms of cutoff frequency, attenuation in dB, and band-stop, test frequency, and gain flatness across a pair of pass-band frequencies.
Then, I selected limits for group delay and step response, in terms of settling time and settling time error band. That selection produced separate designs for a traditional Gaussian filter to 6 and 12 dB, plus equiripple, and Chebyshev filters, in each case with plots of frequency response, group delay, step response, and phase response.
For whichever configuration I selected, WEBENCH presented me with schematic diagrams with recommended component values and National part numbers. It also allowed me to run simulations on each design. The final WEBENCH step, “Build It,” produces a complete BOM and the opportunity to order a kit of all those materials from your distributor.
The eLab Design Center
TI’s tool set, the Analog eLab Design Center, presents a host of tools, most notably the “Pro Series,” which consists of Switcher Pro, for switching power supplies, along with ADC Pro, Clock Pro MDAC-buffer pro, and Filter Pro. There are many more offerings, though, including simulators, videos, and apps data.
TI said Filter Pro provides the ability to adjust passive element tolerances and view response variations, scale passive component values, and view and export filter performance data to Excel. The company also said that the FilterPro active filter design tool allows designers to create and edit active filter designs easily using the Filter Design Wizard. Users can design multiple feedback (MFB), Sallen-Key, low-pass, high-pass, band-pass, and band-stop filters using voltage feedback op amps.
I ran FilterPro to see how it differed from WEBENCH. First, FilterPro runs on the user’s own machine, rather than in the cloud as WEBENCH does. This is a boon and a curse. The boon is that the locally running program is always available. The curse is that you need to make sure you have the current version.
Once FilterPro is loaded and launched, the interface resembles the workbench interface. Users decide what kind of filter to use for the design. I picked a bandpass again (Fig. 1b). Next, users enter the filter specifications into boxes on a graphic representation of the desired frequency-domain response—specifically insertion loss, ripple, the stop-band frequency, and the attenuation within the stop band.
Next, users are asked to specify this filter response as linear, Bessel function, Butterworth, Gaussian to 6 or 12 dB, or Chebyshev. And again, users select a topology.
With that information, the software cranks out schematics and plots of frequency response, phase response, and group delay, all handy on one scrollable screen. A subsequent screen presents numerical values to five significant figures for all of the characteristics in the plots. Subsequent screens provide a BOM and a summary of the characteristics of the filter.
The Analog Devices and TI tools represent the combined engineering talent of a large number of design and apps engineers. In contrast, Intersil’s iSim Active Filter Designer is heavily influenced by senior applications engineer Michael Steffes, who has strong opinions about certain “myths” relating to active filter design. (See “Move Beyond Active Filter Design Mythology To Improve Your Discrete Op-Amp-Based Designs” at www.electronicdesign.com).
Steffes is concerned with the Sallen Key low-pass filter topology, and he’s particularly concerned with common practices related to the way that gain is distributed in multi-stage filters. These concerns show up in the introduction to iSim.
The first steps in filter design using iSim take place in the cloud, but there is also a downloadable component. Intersil said that after using iSim to design a schematic, users can download the offline iSim:PE version of the schematic so they can capture schematics, view waveforms, perform application analysis, and more.
“Many vendor tools provide some filter shape help as an early step in their tools. This is used to arrive at a desired filter order and pole locations to hit a particular ‘skirt’ shape (how fast the cutoff band rolls off). Usually this is specified in terms of stop-band attenuation at a certain frequency above the desired passband,” Steffes said.
“The Active Filter Designer assumes you already know the target shape and/or the approximate order or filter poles you want to implement,” Steffes said. “The tool mainly works on getting the right op amp selected and design implemented in a way that will yield a successful board level implementation.”
For users who need help deciding on the filter shape, Steffes suggests a site called Filter Wiz PRO, which he describes as a “free download that has a lot of filter shape design tools.”
Further steps provide detailed guidance, making the Intersil tool a different experience than the others. What it may come down to, however, is whether the user wants a quick solution with what is essentially a “known-good” solution or a rigorous tutorial.
Linear’s Web site offers LTspice, amplifier simulation and design, filter simulation and design, timing simulation and design, and data converter evaluation software.
LTspice IV is far more than a Spice simulator. It provides for schematic capture and waveform viewing. It’s optimized for the simulation of switching regulators, though, so I didn’t want to compare it with the active filter designers covered here.
On the other hand, the company’s amplifier simulation and design tools comprise a library of the company’s op-amp Spice models for general analog circuit simulations. In addition, the Configurator design tool generates complete schematics for amplifier stages designed with any of the company’s gain selectable amplifiers.
Linear’s Noise program allows users to calculate circuit noise in designs that use the company’s op amps and optimize for a low-noise application. The most remarkable app, FilterCAD, was developed to help users create filters using Linear’s monolithic filter ICs, the LT1567 and 1568. These are not new parts. They’ve been around for a decade and a half, but they remain unique.
The LT1568 is an active-RC filter building block with rail-to-rail inputs and outputs. The internal capacitors of the IC and the gain-bandwidth product of the internal low-noise op amps are trimmed so consistent and repeatable filter responses can be achieved.
Also, the LT1568 is resistor programmed. With a single resistor, users can create a design that provides a pair of matched two-pole Butterworth low-pass filters with unity gain. That’s for starters. By using unequal-valued external resistors, the pair of two-pole sections can create different frequency responses or gains. Alternatively, the two stages can be cascaded to create a single four-pole filter with a programmable response or a band-pass filter. Cutoff frequencies can be as high as 10 MHz.
The LT1567 analog building block is similar. It can be used to build a second-order antialiasing filter with a cutoff frequency up to 5 MHz.
With these ICs, FilterCAD will design low-pass, high-pass, bandpass, and notch filters with Butterworth, Bessel, Chebyshev, elliptic, minimum-Q elliptic, and custom responses. The software shows its age by being configured for 640-by-480 screens, which leaves a lot of blank area on a high-res monitor. But as a quick and dirty way to whip up a filter solution in an industrial-control situation, the combination of IC and software could prove handy.
Maxim’s downloadable EE-Sim Design Generation and Simulation Tool generates an interactive schematic complete with adjustable components based on user input. Detailed input requirements make it possible to create and simulate circuits online.
The user interface is text-driven, presenting a list of filter types and response characteristics based on specific Maxim op amps. After users select a type, the tool presents a schematic and boxes where users can enter the usual filter characteristics. When that input is complete, the filter is designed.
This is followed by a simulation step. Each design iteration can be simulated quickly. The resulting time and frequency-domain plots and graphs are available for analysis in the tool’s waveform viewer, which provides zoom-in capabilities.
IC Design Is Different
What tools do analog IC designers use? How different are they from EDA tools for digital design? The answer depends on what process technologies you’re talking about. Previously, I wrote about tools that were designed to work with foundries’ process technologies (see “Integration, Evolving Markets Cast Analog Designers In New Roles” at www.electronicdesign.com).
For this report, I sought out a company that uses foundries when the results are acceptable, but continues to use its own process technologies when the exceptional is demanded. I also was looking for the opportunity to probe the mind of an analog chip designer who was experienced, yet relatively young, who could articulate what it was like to actually design creative analog circuits in today’s environment.
At Linear Technology, I spoke with Leonard Shtargot (Fig. 3). I began by asking him for an example of product development and the tools that had gone into it. He surprised me by his answer.
Where I had expected him to talk about layout and doping, he spoke about analog design from the standpoint of customer needs. I thought we might be going off-track when he started talking about regulators, but that was his way of getting to a point about band-gap references.
“In large-scale applications, the people who are concerned about power conversion aren’t very much concerned about how much the converter itself consumes. But for applications like backup memory, sensor powering, and so forth, you need high efficiency at very low load currents. The way that Linear originally attacked that was, when the output of the converter was in regulation, we would shut down all the circuitry inside the chip that wasn’t necessary,” he said.
“Consider the subsystems within your car that must stay on when the car is off. The automotive engineers wanted to keep current drain for those elements down to 50 or 100 µA. Initially, that led to a line of products we called burst-mode. But the automotive guys wanted even lower current drain, so they’d use a switching regulator when high power was needed, but then switch to an LDO (low-dropout regulator) for low currents,” he said.
“But that’s really complicated and requires a lot of logic. So we thought about designing a switcher containing a 1-μA reference in which the bandgap runs only in the hundreds of nA.”
Did he do that? Not directly, and the way Linear went about the design speaks volumes about the company’s collaborative approach to chip design.
“Now, we have an internship program with MIT (Massachusetts Institute of Technology). And we thought that would be a good project to give to one of those co-ops. He worked on that for a few years and we wound up hiring him and, ultimately, released a part called the 3971. It has 1.5 to 2 µA of quiescent current, which compared to our former part is a full order of magnitude better,” he said.
Naively, I asked about the tools the student had used. The tools turned out to reside in the brains of more senior engineers. Shtargot said that the best asset is the experience and intuition of folks who’ve done this before.
The student had to start by asking the designers who really understood voltage references and the designers who really knew switching supplies about their experiences with combining those circuits with other circuits, then experimenting to find out what problems turned up, and going back to get more feedback. The problems were very multidisciplinary, so they didn’t lend themselves to tool-based solutions.
“The layouts have to be correct, the stresses in the chip have to be okay, and the substrates have to be connected in a certain way,” Shtargot said. Nobody has built that into a generic tool.
“We do have our own processes,” he said, “which makes it convenient to learn about device properties. For something like this, you do have to handcraft individual devices. Sort of look at one transistor for a day and decide how you want to draw it to do what you want.”
I asked what process node we were talking about, and he said 1.5 µm, noting that while Linear has much smaller geometries, the key to the voltage reference design was being able to analyze each iteration.
“That’s important, because when you do something like this, you have to be able to put it on a probe station and probe it. You can’t just run simulations over and over again,” he said.
“A great deal of it has to be based on decades and decades of experience, because in circuits like this, designers are required to be more like physicists than like somebody who looks at a circuit and decides that it has enough zeroes and ones at the right nodes,” he said.
“For that part, we had to make a band-gap that could run at very low currents. That required transistors for which we had a lot of information about lifetimes. A smaller process might have resulted in a product that was less dependable. And for a reference, small is not the objective, anyway. For a precision thing, bigger is often better.”
That story seemed to be from some time ago, so I asked Shtargot what he was working presently. It turned out to be related. He said that his group was still involved with ICs that exhibit ultra-low quiescent current consumption, which continues to be something that Linear can leverage with its automotive customers.
Since Shtargot had emphasized collaborative design, I asked about his personal experience with playing on various design teams. He emphasized the importance of continuity.
“LTC (Linear) is a really interesting place because a lot of the folks that work here were either part of the original wave of innovation that created the first analog ICs or the next generation. So seeing how their minds work about designing integrated circuits and solving problems is really valuable,” he said.
“As an undergraduate engineering student, I learned analysis techniques, and how existing circuits work, but there wasn’t an emphasis on creating new circuits. Working at Linear, I learned the missing 70% or 80% about creating something new,” he said.
“Analysis is certainly important, but doing something new requires the human mind to build abstractions in terms of the various building blocks, abstractions that are simple enough so you can synthesize them to do what you want. That’s a tool that you can only learn from people who’ve been doing this for decades. The people who did these things originally looked at circuits from their physical embodiment.”
I asked him for an example.
“In switchers, where I work, you have very high di/dts, and when you have that, a lot of things start ringing. You can’t ignore those fundamentals because either the IC just won’t work or it will generate excessive EMI (electromagnetic interference),” he said.
“You learn how to deal with that when you get the microscope and the small soldering iron and deal with the physical reality. Compared to digital design, with analog, you are flying very close to the limits of the physics, because you’re manipulating energy, rather than information. And you can do that, where the digital designers can, because semiconductors for reasonable power levels are quite big, physically,” he said.
Shtargot didn’t hold much hope for a future generation of analog EDA tools.
“Another thing you can actually learn from more experienced designers is what I’d call intuition. Current EDA tools don’t contain enough ‘intelligence’ to make an analog circuit work right the first time, regardless of the number of simulations you run,” he said.
“Spice works well if your models are good, but characteristics change with how the device is drawn, or with relative temperatures. That’s hard to model. Essentially, the problem is that if you tried to build a generalized CAD system that could capture any variant of an analog circuit that you’d want to design, it would take much longer than to go and talk to experienced people who can pass their knowledge along and then just build the circuits. Maybe one day programming techniques will get better.”
Another aspect of using the brains of experienced engineers to guide designs, rather than relying on canned software, is the ability to redefine the problem in terms of what the silicon can do instead of trying to force the silicon to do things it isn’t happy doing.
“With analog design, a big part of the challenge is defining the problem, defining what you really want,” Shtargot said.
“One of the things that we learn from folks here who have a long history is that, when you visit a customer, you have to ignore what the customer said he wants, and listen to his problem. It’s often better to apply the fundamentals to solve the problem, rather than try to force an existing part into a role it wasn’t intended for.”
The focus on physics brought up a point about the security of intellectual property (IP) when it’s based on that depth of knowledge, rather than on layout. Shtargot had stories about people trying to reverse engineer products—how they could duplicate the circuit elements and the layout, but when it came to power up the part, it didn’t work.
Specifically with regard to using foundry processes, Shtargot expressed concern about the foundries’ opaqueness about their process technologies and their reluctance to do any tweaking.
“At LTC, not only do we get to tweak the processes, more important is that we know exactly how it’s made. We can combine layers in ways one couldn’t have in the foundry’s process. You don’t know what the details are. You don’t know the thermal cycles, how deep some of the junctions are, the interface metal, and so on. Or, sometimes they’ll tell you, sometimes they won’t,” he said.
“And then suppose you decide to build something on the foundry’s process. They may later decide to buy a new implanter or other piece of equipment, and that changes the performance of your analog circuit. The matching may not be quite as good, the temp coefficients may be a little different, and you’re not aware of it,” he said.
“Essentially, dealing with that means that you have to have an engineer full time in their fab, keeping an eye on things from batch to batch.”