"Once you have accomplished the impossible, your employer will make it a regular part of your job." The semiconductor industry, in my experience, makes a truism of that observation by Lai See, one of the characters invented by humorist Nuri Vittachi, who wrote regularly for the South China Morning Post in Hong Kong until that city was absorbed by the People's Republic of China.
Anybody who has spent time on the Route 128 circle in Massachusetts will remember some seemingly impossible tasks in data converter design: resolving the terrible paradoxes among bit resolution, speed and power consumption. Twenty years ago you could get a 16-bit data converter that sampled at 1-MHz rates, but chances are it was a hybrid circuit on a ceramic substrate, packaged in a gold tub and custom built for the military. Bernie Gordon, Analogic Corp.'s famous president and chief technologist, was so convinced of the technology limitations that he declared one year at the International Solid State Circuits Conference that a monolithic 12-bit converter was a physical impossibility.
Even when single-chip A/D converters with 12-bit successive approximation registers went into production, their reference D/A converters were based on an R/2R resistor ladder that had to be laser trimmed for accuracy. (The least significant bit, or LSB, in a 12-bit string is 1/4,096 - 1/2 to the 12th - of the most significant bit, or MSB. The LSB in a 16-bit string is 1/65,535 of the MSB.) This meant that every device needed parametric testing, which made its cost at least $20.
Semiconductor technology, especially high-density CMOS, has changed much of this. There are single-chip 16-bit devices that will sample at 1-MHz or more. But you know the tradeoffs among resolution, speed and power consumption are still there. It's just that the boundaries have been pushed out. You can get a data converter from Maxim Integrated Products that samples at rates up to 1 GHz, but it will only resolve 8 bits. Conversely, you can get a dc-accurate 24-bit converter from the Industrial Products Group at Cirrus Logic's Crystal Semiconductor Division, but its primary application is weigh scales that take (maybe) two readings a second.
Let me point to some interesting points on the trend line: In the high-resolution area, sigma-delta technology (or, more appropriately, delta-sigma technology) is making it possible to build audio data converters with 24-bit resolution. These are not dc-accurate parts, remember; you do not want to use them for precision measurement and control applications. But they lend themselves well to audio. They use digital noise-shaping techniques to construct something that resembles a grossly oversampled signal and decimation filters to extract in bit form a very accurate facsimile of the digitized audio signal (by observing the changing sums). Thus, manufacturers like Crystal can effect a 120-dB dynamic range and a 96-kHz sampling rate. (As I've written before, this kind of audio dynamic range is equivalent to recording caterpillar footsteps on an airport runway with an SST taking off overhead. It's almost beyond the range of human hearing).
Interestingly, Crystal's devices - along with those from competitors like AKM Semiconductor, Analog Devices, National Semiconductor, Philips and SigmaTel - will make this technology increasingly accessible to consumers. D/A converters using variations on delta-sigma technology already are a staple in audio CD and DVD players. We expect to see it in digital camcorders and consumer recording equipment as well.
In the high-speed area -40 to 60 MHz - manufacturers are trying to increase the resolution of the A/D converters used for signal capture in cellular base stations. The direct-conversion methodology hopes to eliminate a lot of radio frequency (RF) and intermediate frequency (IF) tuning and mixing components by capturing a wide swath of the radio spectrum, converting the captured carrier and modulation signal directly to digital data and using a digital signal processor (DSP) to extract the digitized voice or data components. Currently, base station manufacturers are working with 12-bit converters that sample at 65-MHz rates, from companies like Analog Devices, Burr-Brown and National Semiconductor. The base station makers undoubtedly would like higher speeds and bit resolutions, but the tradeoffs suggest one or the other. For example, Signal Processing Technologies in Colorado Springs, has a device that will sample at 100 MHz, but its resolution is only 10 bits.
I think the bigger push is on higher resolution - 14-bit devices - that will sample in the 40-to-60-MHz range. The higher dynamic range will allow manufacturers of cellular base stations to capture smaller distant signals in the presence of strong close ones. Both Analog Devices and Burr-Brown have announced products in this area, and are racing to get the devices production worthy even as they explore higher resolutions and speeds.
But it must be said that this is not yet a low-power CMOS technology. These pipelined devices are not self-clocking: you have to crank them pretty hard to capture a 30-MHz RF swath with Nyquist criteria (that is, 2x oversampling). So this kind of direct-conversion technology does not readily lend itself to portable handheld devices. I wouldn't say manufacturers who tell you they've got a direct-conversion receiver for cellular handsets - or, interestingly, for 2.45-GHz Bluetooth - are blowing smoke. A number of these startups are well funded, meaning the investment community has faith in their ability to deliver. But I believe that we need to go through two CMOS generations (three to five years) before direct conversion will be practical for handhelds. In the interim, cellular handset and Bluetooth receivers will reflect conventional tuning, mixing and downconversion architectures - with a zillion pesky passives.
This doesn't mean that some manufacturer of low-power direct conversion chip sets won't be out to show me up. When Precision Monolithics (now Analog Devices' Santa Clara Division) pulled off Mission Impossible with one of the industry's first monolithic 12-bit D/A converters, the part's designers offered a clear message to Analogic's Bernie Gordon and other conservatives who would have said this kind of integration was impossible. Into the metal mask of the chip they implanted the initials "B.G." and a heart shape with a bite taken from it. To those who thought this stuff was impossible the message was, "Eat your heart out." Look what the semiconductor industry does every day: it builds New York City on the head of pin and still makes money doing it!
If you find this material interesting (please give me feedback, either e-mail me direct or post a comment under "Steve wants to know..."), I'll give you my 15,000-foot view of op amps and voltage regulators as well. I can't guarantee, though, that I'll always have an interesting song lyric or soothsayer's wisdom to share.