The role of the analog designer—and, to some extent, the power-supply designer—has been evolving over the past decade or so. Recently, the trend has been accelerating, driven by ever higher semiconductor integration levels and global changes in product manufacturing.
Integration has been moving upstream from the processor toward a variety of analog peripherals, and manufacturing has been moving to China. The longer version of the story, though, is interesting and instructive.
Consider the evolving role of analog designers. At one time, these engineers were expected to be masters of the arcane art of interconnecting an assortment of more or less single-function components. Long ago, they worked with passives and tubes, then transistors, then operational amplifiers and comparators, then simple timing ICs, phase-locked loops (PLLs), analog-to-digital converters (ADCs), digital-to-analog converters (DACs), and so forth.
These designers were engaged with driving analog radios and TVs, oscilloscopes and X-ray machines, and radar sets and bombsights, most of which interacted with the real world in terms of analog signals. More recently, the output became digital to facilitate interaction with microcontrollers or DSPs, which were the domain of the digital designers (see “An Early Experience With Mixed-Signal Design Disciplines”).
Today, it’s different. When mixed-signal chipmakers announce new products, many of these devices, perhaps most of them, are application-specific. They also comprise the entire signal chain, from the sensor inputs that collect analog data in one form or another and turn it into voltage levels through signal conditioning to digitization.
Often, the output is preconfigured for input to one of the chipmaker’s microcontrollers or DSPs. In fact, a complete reference design is usually available. All that is left for the end-product manufacturer to do is to decide which features to implement at which price points and to create a package design that will wrap around the electronics.
Forces for Change
This chipmaker business model seems to be common around the world. The only difference is that in Europe, Japan, and Korea, where IC makers may be part of a larger, vertically integrated organization, the prime customers for new chips may be internal, while for North American chip companies, the customers tend to be manufacturers in China.
For analog design engineers, this change in product integration has altered the career landscape. Except for engineers involved in industrial control and, to some extent, in mil-aero and telecom companies, the jobs that once involved bench design have evaporated.
All of the companies that I have spoken with agree that analog design opportunities have migrated inside the chip companies, where analog engineers may be mixed-signal chip designers or application engineers, either inside or in the field.
The story of how this state of affairs came to pass takes some telling. Obviously, the relentless shrinking of IC design rules, generation by generation, enabled greater integration. But that same relentless shrinking created enormous headaches as operating and allowable input voltages shrink along with feature size.
Refining analog chip design in the analog domain alone probably stopped around the 0.35-µm generation. Ever since then, it has been necessary to accommodate what are essentially noise problems with tricks in the digital domain.
In fact, that design-rule shrinkage has been one of the drivers toward mixed-signal integration. Interfacing a signal-conditioning amp to an ADC no longer can be left to customer engineers, who aren’t intimately familiar with the quirks of the process technology and the often proprietary techniques that have been implemented at all stages of the signal chain.
Chipmakers that don’t want customers complaining that their products can’t achieve their datasheet characteristics in real-world designs must hide their interfaces inside the chip package. Even then, it’s unlikely that the performance of the final product is going to be improved by a customer’s tinkering with the layout developed at the factory.
Hence, we find reference designs complete down to Gerber plots and armies of field application engineers to explain to customers why they really don’t want to save half a cent per board by using cheaper capacitors.
The notion of pecking a design to death by chipping away at bill of materials costs brings up the other major reason for the move to highly integrated mixed-signal IC products. This part of the puzzle relates to the companies that are buying the new types of mixed-signal products and turning them into consumer products for you and me and the emerging middle class in the Peoples’ Republic—the OEMs and ODMs.
It’s useful to understand those acronyms—original equipment manufacturers (OEMs) and original design manufacturers (ODMs). I cannot find a definitive explanation on the Web that I agree with, so I will propose one that people in the chip business have told me.
The short version is that it’s a tiered situation. OEMs make products that have brand names for companies that sell those brands. ODMs make subassemblies that multiple companies may slap a GUI on and enclose in a case and sell under any number of off-brand names.
The OEM designation is an extension of the OEM concept in the automotive business. When Ford or Toyota sells you or me a windshield-wiper motor, it was probably made by an outside company that built it to Ford or Toyota specs.
The motor probably also underwent some kind of acceptance inspection before it entered Ford’s or Toyota’s depot-to-dealer chain. The Ford or Toyota brand, and the fact that the motor comes from the car company’s parts department, gives us confidence and supports a higher purchase price than a product with similar dimensions that we might purchase from a catalog.
In the case of electronic products from a manufacturer in China, who may have purchased branding rights to a once well known North American brand name, it works in a similar fashion. The products the company with the brand puts its name on are unique, and the company has been working, possibly for a decade or more, on establishing a reputation for its brand through excellent performance, intrinsic quality, and customer support.
In China, this kind of company tends to target the overseas market. That’s not so good when the overseas market is having a recession, as it is now, so that it can’t support the customary short consumer-product life cycles. (We’re supposed to buy a new TV every three years.)
Meanwhile, there’s a growing middle class in China. There are potentially more people in this group than there are anywhere else on the planet, and they generally don’t have any nifty consumer products, so they’re eager to buy now.
It’s also necessary to keep in mind that they don’t have credit cards. So while they can’t spend as much as Western consumers, they are less likely to feed economic bubbles, which can be a good thing if, like China, you’re only just evolving from a strict Marxist economy.
That’s where the ODMs enter the picture. If a chip company’s reference design has enough feature hooks to support a common design with a range of price points related to how many features are implemented, and if it can be offered in different configurations by multiple domestic brands, and if the ODM can get it into manufacturing quickly, with minimum non-recurring engineering (NRE) costs, then that product is a winner.
Or it’s a winner if it will support the end users’ price/performance expectations. For example, a long time ago, I wrote about air conditioners for the Japanese and Chinese markets (see “Air Conditioner Chip Set Is Way Cool” at www.electronicdesign.com). The issue was the kind of motor control needed for different kinds of compressors and the tradeoffs between price and quietness. Japanese consumers would pay extra for a quieter unit. Chinese consumers just wanted to stay cool.
Beyond economics, there is a host of other cultural considerations. When you have vast sections of a country with few doctors and hospitals, relatively inexpensive medical instruments that can be used by relatively untrained medical technicians will make a large difference in overall health care. Thus, you will find ODMs making not only cheap laptops, air conditioners, and cameras, but basic ultrasound machines as well (see “New Technology Treats Medical Needs in Developing Countries” at www.electronicdesign.com).
OEM or ODM, the business model doesn’t work if it requires high NRE at the manufacturer. That’s another factor in favor of concentrating the more intense analog design efforts at the semiconductor manufacturer, rather than following the older model of developing a handful of ICs that were capable of high performance individually, on the bench and, in effect, “throwing them over the wall” to the end-product designer.
Avoiding that disconnect plays into an aspect of recent Chinese history as well. Any time I talk to senior engineers at semiconductor companies about where future generations of analog designers are going to come from, they invariably tell me that new college grads need at least five years of mentoring by senior designers before they really come into their own.
The young students who have made it into engineering schools in China are unquestionably brilliant and hardworking. But when they graduate, they face something of a vacuum in senior mentors. I’m told that this is partially mitigated by the return of retired engineers who have worked abroad for decades, but it won’t be completely eliminated until the more recent graduates have gone through their own maturing process.
That doesn’t mean the young Chinese engineers have been standing still. According to several sources, they rely heavily on their own social media for mutual support on technical issues. The primary channel is 21IC (www.21ic.com/).
So far, this discussion has concerned traditional semiconductor companies and their customers. But the trend toward high mixed-signal integration and providing complete productization through reference designs extends to fabless startups as well.
To cite one example, Samplify evolved from an intellectual property (IP) supplier to a vendor of ADCs to the source of a complete product design for portable ultrasound devices (see “The Mind Of An Entrepreneur” at www.electronicdesign.com). Samplify’s approach is optimized for dealing with ODMs, but that’s not the only route that fabless companies follow. Wolfson Microelectronics says that it seeks to get its codecs onto the reference design boards of larger companies, allowing it to sell into OEMs.
When discussing this trend, it’s hard not to dwell on the Far East because of the manufacturing volume that part of the world generates in the form of consumer products. Yet higher integration and pre-optimized designs are starting to appear in the Western industrial control market as well.
By definition, industrial control is regional, focused on specific industries, and the number of ICs that any single project requires is low, compared to what goes into consumer products. Nevertheless in the aggregate, it’s big business for distributors.
The problem is that, with the semiconductor companies absorbing the greater number of analog new college graduates and putting them to work as chip designers and applications engineers, there’s a scarcity of new analog designers available for industrial-control development work, either inside big companies or in small design organizations. Along with this, the senior designers are getting old and retiring, creating a loss at one end of the experience curve that aggravates the dwindling supply at the other.
In recognition of this, some semiconductor companies have begun to adapt. A good example is National Semiconductor. One aspect of National that will surely survive its acquisition by Texas Instruments is its portfolio of WEBENCH online design tools.
Historically, these tools have been aimed at creating designs out of ICs that are, functionally, relatively primitive, compared to those being considered in this article. But that changed recently with the introduction of signal-conditioning products customized for specific kinds of sensors.
Complementing the new hardware is a new WEBENCH implementation that customizes the entire signal chain (see “Chips Implement Sensor Signal Chains From Nanoamps To Bits” at www.electronicdesign.com). One example of the sensor products is the LMP91000, a programmable analog front end (AFE) for electrochemical sensing applications, described at www.national.com/pf/LM/LMP91000.html#Overview.
The makers of power-management ICs have been following a similar path to the makers of mixed-signal ICs ever since the dawn of digital control in switching regulators. This is especially true for applications that use the Intermediate Bus Architecture (IBA), where closing the control loop in the analog domain has always been perceived as an analog black art.
Thus, chipmakers such as Power-One and National Semiconductor supported their “digital” power products with GUIs that took all of the bench work and much of the analysis out of the design process. For the most part, these products were dc-dc switching regulators. But Vicor has recently introduced a site called PowerBench (www.vicr.com/cms/home/technical_resources/powerbench) that supports designs from simple buck and boost regulators to complete, multi-output ac-dc supplies.
Several semiconductor companies have made some recent announcements that illustrate these kinds of mixed-signal chips. For example, last March, Texas Instruments announced 16- and 14-bit, two-channel, simultaneous-sampling successive approximation register (SAR) ADCs with two independently controlled internal references for simplified system-level design (Fig. 1).
The 16-bit ADS8363, 14-bit ADS7263, and 12-bit ADS7223 provide twice the per-channel throughput of similar ADCs, supporting speeds up to 1 Msample/s. These ADCs target industrial control rather than consumer or medical applications. Typical applications include motor control, power quality measurement, power automation, and solar and wind-power inverters.
The two independently controlled 2.5-V references were integrated to enable separate ADC gain calibration and the implementation of a programmable gain amplifier (PGA). Alternatively, this feature also reduces the need for external signal conditioning when different gain levels are required for each ADC.
New products from Linear Technology frequently target automotive applications. In terms of high integration, Linear’s LTC6803 second-generation high-voltage battery monitor for hybrid electric vehicles (HEVs), electric vehicles (EVs), and other high-voltage, high-performance battery systems deals with the problem of high voltages across a series battery stack (Fig. 2).
Multiple LTC6803s can be stacked in series without optocouplers or isolators, permitting precision voltage monitoring of every cell in the array. Each LTC6803 battery measurement IC includes a 12-bit ADC, a precision voltage reference, a high-voltage input multiplexer, and a serial interface. A single LTC6803 can measure up to 12 individual battery cells in series.
Cell voltages from –300 mV to 5 V can be measured enabling the LTC6803 to monitor many different battery chemistries, as well as supercapacitors. In addition to the precision measurement, each cell is monitored for undervoltage and overvoltage conditions. An associated MOSFET is available to discharge overcharged cells. There is also a 5-V regulator, a temperature sensor, GPIO lines, and thermistor inputs.
The unique requirements associated with measuring vehicle-side battery arrays led to some equally unique solutions. For example, for long-term battery-pack storage, the current consumed by an integrated battery management system could unbalance the cells. Linear’s response is a standby mode that draws less than 12 µA.
Alternatively, since the power input of the LTC6803 is isolated from the stack, the LTC6803 can draw current from an independent source, in which case the current draw on the pack is less than 1 µA. Beyond that, the chip has to meet automotive temperature, environmental, and safety standards. All this costs $9.95 per chip in 1000-unit quantities.
In January, Analog Devices introduced its first product in a projected line of AFE chips for electrocardiogram (ECG) systems (Fig. 3). An ECG system measures and records the electrical activity of a human heart, making it possible to diagnose and analyze birth defects, arrhythmias, problems with heart valves, and lack of blood flow to the heart muscle.
The ADAS1000 ECG AFE eliminates approximately 50 components from the signal chain in a five-electrode ECG system. The device also incorporates pacemaker pulse detection and respiration measurement. It is something of a one-size-fits-all device in that it can be configured to optimize noise performance, power, or data rate. Those are the different design targets for home, ambulatory, and clinical ECG systems.
Naturally, for ODMs and OEMs, there is plenty of design support. An evaluation board includes the AFE, power supplies, and control and interface options. And since this is only the analog front end, Analog Devices bundles it with its Blackfin DSP, a single-chip USB isolator, a four-channel digital isolator, a 5-kV dc-dc converter, a 433-, 868-, and 915-MHz industrial, scientific, and medical (ISM) band transceiver, and various dc-dc regulators.
TI also has an AFE family for portable ECGs. The latest, the 24-bit ADS1298R, includes respiration detection, integrating more than 40 discrete components that would otherwise be required (Fig. 4). The chip also uses up to 95% less power than discrete implementations. On-chip are eight ADCs, eight PGAs, and eight active filters along with a pacemaker detection interface, lead-off detection, a voltage reference, right-leg drive (avoids common-mode pickup), and more.
In the reference design, the chip is paired with an ultra-low-power DSP. Demonstrating the amount of design support that chip companies include in these targeted mixed-signal chips, the reference design also generates oscilloscope, fast Fourier transform (FFT), and histogram displays. In 1000-unit quantities, the eight-channel ADS1298R costs $25.95. Six- and four-channel versions go down to $13.95.
Sometimes, the degree of integration can include micro-electromechanical systems (MEMS). In January, Analog Devices introduced the ADMP441 high-performance MEMS microphone with an I2S (Inter-IC Sound) digital output. Production quantities became available in June.
The company had introduced other microphones in its iMEMS family. But the new mike eliminates signal conditioning and digitization, providing 24-bit serial data output directly (Fig. 5). Specs include 100-Hz to 15-kHz frequency response, 61-dBA signal-to-noise ratio (SNR), and 80-dBFS power-supply-rejection ratio (PSRR). Its package measures 4.72 by 3.76 by 1.00 mm. At introduction, pricing was $2.38 per unit in 1000-unit quantities.