The Other Technology — An Outlook

This is the text accompanying a presentation by Bob Swanson at an analysts’ conference sponsored by DataQuest in 1983. There were slides presented, as is apparent in the text, but I don’t have good enough copies to include them here.

Over the past several years you’ve probably heard and read a lot about the world of VLSI and its push to put more and more on a single chip. The other technology my talk refers to is the technology of linear integrated circuits. In order to get a better understanding of the linear business, it might be helpful to compare it to the digital side of the industry, the side you’ve heard about, the microprocessor/memory/VLSI side.

VLSI is the business that works at the very limit of the processing technology, not because they want to but because they have to.

For starters, the basic linear wafer fabrication process is not high density, and relatively less capital intensive. Our processing technology is less vulnerable to the obsolescence created by the rapid advance in fab equipment that’s required in the VLSI sector. This is especially true in the photo masking area. (see figure 1)

The first slide shows a rough comparison of the masking equipment cost for the high-density processing of VLSI wafers, the productivity, and typical net die per wafer. This illuminates the dramatic capital equipment cost difference that exists between linear and digital VLSI on a per die basis.

For Linear, the object is not to make things smaller. Smaller typically is bad. Smaller features mean more noise. The bigger the feature, the more precise it can be and the easier it becomes to precisely match one to another. Matching features is a critical part of the linear circuit design task.

Linear also means high voltage; typically 30–50 volt operation is required as opposed to the 5 volt world of digital. Because higher voltage is needed to serve traditional market areas and because higher voltage means more precision — because signal-to-noise ratio also improves with higher voltage — we in the linear business are often precluded from shrinking features and spacings.

In fact, we often find ourselves spacing things out, making junctions deeper, etc., which makes die features bigger.

The linear circuit designers claim they optimize device characteristics rather than optimize device size.

If you look at past trends in an admittedly simplistic way, digital design made things better by adding more and more of the same on a single chip. Linear, on the other hand, was constantly making things more precise by refining what they had and creating new functions.

The next slide shows a logic gate, a basic digital device, as well as an operational amplifier, a basic linear device. Both have two inputs, both have a single output, but from there the similarities disappear.

The next slide shows how digital integration takes place. There is only one way to hook up a NAND gate. Because of this, CAD can hook up many of them to create MSI, LSI, VLSI, limited only by the realities of wafer processing capability. This business is CAD-intensive and it’s where the computer can out-perform the human being. On the analog side you can’t make better linear circuits by adding more and more of the same, e.g., ten circa-1970 op-amps don’t equal one circa-1980 op-amp. The progress on this side of the business is achieved by improving the performance of individual op-amps. Unlike the gate that could be hooked up in only one way, the op-amp can be hooked up in an infinite number of ways, primarily dictated by the use of external components, such as resistors and capacitors, and the creativity of the user. As the attempt is made to integrate more of the linear portions of the system, the IC producer runs into the problem of integrating these many external devices into the larger IC solution, i.e., it’s not easy.

The linear circuit layout (mask design) is still very human-dependent and still a long way from automated computer layout. Some of the reasons are that the linear mask designer must not only optimize space efficiency but also the electrical aspects of the design involving a knowledge of the thermal and crystalline properties of the silicon. Linear circuits are not iterative, i.e., few elements are used over and over again. Each part of the circuit is different and the correct interaction of these different sections is a major task. This interface in a digital IC is much less of an issue.

Operational amplifiers make up an important and fundamental class of linear devices. Two significant parameters of op-amps are offset voltage and bias current. These are both error terms, so the lower the value, the better the device. Noise is another important parameter. The next three slides show the progress made over the years at improving these basic performance parameters.

The bettering of VOS since the mid-60s represents a 300 times improvement. Bias current equals 100,000 times improvement and noise a five times improvement. Another important area of progress is the fact that Linear Technology Corporation has introduced op-amps this fall that for the first time offer state-of-the-art specs for many of the important parameters on a single chip. Previously, one op-amp had the best offset voltage, another the best bias current, and so forth. It was not possible in the past to create one device that was best on all parameters. We’re beginning to change all that at LTC.

If you look at some more of the history of how the linear IC business got started, you begin to understand how linear designers adapt and manipulate what exists.

Back in the mid-60s, the digital bipolar process was developed using only NPN transistors. Along came a linear circuit designer named Bob Widlar who observed that this process could also be useful to linear, especially if PNPs could be added. They were and Widlar almost singlehandedly started what is today a $2 billion market.

There are two points to make here:

1. the linear IC business has been fashioned mostly by circuit design innovation
2. Even the process innovations were driven by an individual circuit designer or a small group at most, not an obvious industry-wide push by large teams of engineers.

Linear is a business of individual contribution and it’s a business where a few can make an enormous contribution.

Some other characteristics that distinguish linear from other segments of the IC business are:

• Our business is fragmented; it’s a small volume, large list of device types produced from a wide variety of processes in wafer fab and sold to hundreds of thousands of customers.

APPLICATIONS

Let me show you how linear circuits are used in conjunction with other devices or end products you’re familiar with.

This next slide shows some typical applications of linear ICs in a microprocessor-based system. This is a simple industrial process control application where the pressure and temperature of a pump must be controlled. As marvelous as the microprocessor is, it can speak in digital language only, highs or lows, off/on, ones/zeros. It can’t understand the real-world parameters such as temperature and pressure.

Linear ICs must process, shape, condition, and finally, convert linear signals into a format that the microprocessor can understand. So if the microprocessor and memory chip are the brains of electronic systems, then the linear ICs are the sensory, the nerve, the muscle part of the system. They are the eyes, the ears, and hands of many electrical systems.

From a dollar standpoint, a typical system like this contains more linear dollars than the microprocessor chip itself.

Personal Computer

The next slide shows a personal computer arrangement. Now even I think of a personal computer as a digital machine, with digital in through the keyboard to the digital computer inside the machine. But a close look reveals lots of linear:

• Power supply circuits in the computer itself
• Op-amps, comparators, voltage regulators, and other linear circuits in the disk drive
• More op-amps, comparators, regulators, and driver circuits in the printer
• and so forth

Disk Drive

If you look at the disk drive part of the personal computer setup you’re talking about another whole industry. The disk drive in the personal computer includes some linear. But when you look at the more sophisticated drives, you’ll find that the accurate positioning of the head on the disk and the recovery of the data from the disk both rely on linear ICs.

Scale

Another system that you’ve all been exposed to is the supermarket scale hooked up to an electronic cash register. In this application, linear ICs must provide data that is 1 part in 2,000 absolute accuracy (0–20 lbs +/– 0.01 lb). Although the cash register’s main function is digital, it relies on linear circuits to amplify the low-level signal from the weighing platform, as well as convert the linear weight data into a digital format suitable for the computer.

CAT Scanner

This slide shows a medical application that is linear IC intensive. In a CAT scanner, the donut-shaped assembly rotates about the patient. As it rotates, the amount of x-ray energy presented to each detector will vary depending upon the density of tissue being examined. The computer in combination with the instrument’s rotational motion executes a complex analysis allowing a three-dimensional representation to appear on the display. Although the instrument uses a computer to generate its output, it relies on linear ICs to accurately convert the x-ray detectors’ output to digital signals. As many as 1,000 detectors may be used and each detector has an amplifier and a high-performance one at that.

Pervasiveness is an over-used word in this business, but I can’t think of a better one to describe the almost endless applications for linear ICs.

THE LINEAR MARKET

The linear portion of the total semiconductor industry is a steadily growing sector. This slide shows the growth enjoyed by U.S. suppliers as tracked by SIA from 1976 to 1981.

Both SIA and Dataquest data reveal that from 1974 through 1982, the linear market compounded at an annual 22% growth rate despite two very severe recessions that occurred during this timeframe.

The next slide shows, a now old, SIA forecast for 1983 through 1985.

The 1982 data represent actual shipments.

Linear Technology extended the growth (forecast) through 1990. By 1990, the linear market served by SIA companies could actually reach 7–8 billion dollars which would make it as large as the entire IC market was in 1980.

While I have no way of knowing for sure that this market will continue to grow at greater than 20% per year, at the same time I have no reason to doubt it when you consider that linear circuits are found in every electronic system you can name from video games to sophisticated missile systems. The point is that the linear business is today very big and it’s getting bigger.

TRENDS

The first trend I’d like to describe is in the area of power. The power trend is going in two opposite directions.

Because of the pervasive use of digital circuits that in their aggregate create the need for more current which typically means more power (and also because of the push to make these digital go faster) linear circuits used in the same end applications such as voltage regulators will need to supply higher currents.

At the other end, digital circuits like large CMOS RAMs that consume micro-watts of power in their standby mode, or space projects or battery-operated systems or wherever only limited power is available, the need is created for micro-power linear ICs.

Also, because another trend is more integration of more linear systems on a chip, this direction will create the need for lower power linear elements, since today linear circuits typically use more power than do many of the newer digital CMOS circuits.

Today standard linear circuits work in milli-watts. In the future, a growing number will have to work in micro-watts.

MORE SYSTEMS ON A CHIP

More systems on a chip is an important subject these days. Systems often require a combination of digital and linear functions. The linear world has been watching the digital side of the business attempt to integrate what we think are linear elements on-board the microprocessor. Intel’s 8022 and 2920 are good examples. Digital VLSI technology has progressed to the point of doing large digital systems on a single chip. There has been some demand to incorporate analog or linear functions on the same VLSI chip. This has been done with limited success for a couple of reasons. The quality of the linear function available on the digital fab process is low and the expertise required to do the linear function, for the most part, has not been available to the digital company. Notwithstanding these problems, the trend is real and will continue. As the digital IC becomes larger, it’s possible via CAD techniques to incorporate more linear commodity-class functions. The next slide shows our perception of the degree of combination feasible today.

It is possible today to add linear to the VLSI chip but the performance represents a function generally lower than what would be the commodity end of the present linear IC component business.

Interconnecting a large number of medium- or even low-performance linear cells is still not trivial. Again, unlike digital, putting together a number of low-performance linear cells yields an even lower performance system, since linear errors tend to accumulate rather than cancel.

A trend that may not be all that apparent to the rest of the world is the fact that linear suppliers are integrating more and more pieces of digital circuitry as well as more integration of linear elements onto one chip. The next slide shows our perception of how much easier it is to move digital to the linear chip. Experience has shown that the addition of standard logic on a linear die is relatively trivial compared to adding linear to the digital chip. ROMs and RAMs are being used by the linear designer as an inexpensive way to create more precision on ICs such as DACs and A/Ds.

Microprocessors have rules you must observe in order to talk to them and a true microprocessor-compatible A/D converter contains a lot of digital circuitry on the linear chip.

Let me develop the argument further. (next slide) Two to three years ago, it was really trendy to talk about putting the A/D converter onboard the microprocessor. This is a solution; it can be done. But the question is do you really want to? In a simple system where the major blocks are the A/D converter and the microprocessor, the combining of these two on the same chip can make a lot of sense. This is where a dedicated IC makes economic sense. Putting the multiplexer on the microprocessor seems always to waste precious pins on the processor. In the real world, you’re always looking to the future. Most systems evolve, expand, want to improve at some point in time.

The next slide shows another fundamental approach that we think is better for many reasons. Combining the MUX with a serial I/O A/D converter provides maximum flexibility to the user and leads to a universal subsystem. This combination is a more natural one.

The MUX and A/D are both linear functions:

• Therefore the design mentality is compatible
• The wafer fab process is compatible
• The testing skills are compatible, and
• It doesn’t burden the processor with the A/D

As the fab process on the linear side changes to improve circuit performance, the MUX and A/D benefit. However, as the fab process improves for the microprocessor, it probably is being optimized in a direction opposite to where linear wants to take the development.

Further natural integration of linear I/O functions would include op-amps, references, and filters to the block shown here. The serial I/O approach can be very powerful. This approach can lead to a standard à la RS232 for interfacing all kinds of combined linear blocks to the microprocessor.

I can’t leave this subject before commenting on how arrays and standard cells apply to linear. I’m asked that question all the time. My conclusion is that there is not a real good analogy in linear for the standard cell, gate array approach we see today in the digital world. If there are high-performance linear cells, they are today’s standard linear IC products and therefore available only in the linear company. Improved CAD will assist in the near future to integrate various linear cells that we see today as individual ICs. The task, however, seems more delicate than simply metal-masking a digital gate array to achieve circuit hook-up. Linear blocks can be connected in many ways and by designing linear LSI we tend to dedicate the linear system and although high-performance linear LSI will happen, the move will be slow. We at Linear Technology are not picking our IC products haphazardly. There is a definite pattern to select individual ICs that can make up a library of useful cells for possible future integration.

MORE MIXED PROCESSES

This next slide shows some basic wafer fab technologies available to the linear IC manufacturer. There are basic bipolar and CMOS processes and a variety of adjunct processes. This adjunct variety is so named because they can be combined with either bipolar or CMOS.

Although most if not all of these processes have been used, more mixing will result in the future to achieve unique performance.

The bipolar process and all of its variations is the workhorse of linear. The estimate is that some 80–85% of linear sales come from bipolar products. However, anyone who is serious about the linear business must be into CMOS technology, because as the market doubles or triples between now and 1990, a substantial portion of that growth will require CMOS. As most of you know, CMOS is already a major technology in the memory business. However, this is not the CMOS needed for linear. In order for CMOS to become more useful to the linear designer, linear process people will need to enhance it as they did the original bipolar process to make it: higher voltage, less noisy, provide good capacitors, and in general make it more rugged so it can survive in the hostile real-world of linear.

MORE PROTECTION FEATURES

Intel has warned the microprocessor industry that software know-how or lack of it will limit the growth of the microprocessor market unless something is done about it.

The linear world has a shortage to contend with also. Fewer and fewer linear trained engineers are being produced by the colleges today. A lopsided percentage of recent graduates have been listening to Intel and speak digital only.

But because many of them will get the job of not only designing the larger digital portion of the system but also the linear portion, the linear companies must design their circuits with extreme “ease in use” as an attribute and at the same time try to make them “blow-up” proof.

The voltage regulator is a good example of mission accomplished:

• First of all, there are only three leads to hook up
• If you short-circuit the output, it’s protected
• If you over-voltage it, it’s still protected
• If you run it too hot it will shut itself down

More and more circuits will incorporate protection features to not only save the linear circuit from misapplication but also the expensive digital circuit that the linear circuit is often called upon to protect in the system.

There are many other important trends in linear, but not enough time to cover them all.

Let me wrap up by saying that linear — the other technology — represents a large robust business, one that will grow as the digital side grows. Linear will begin to move into the LSI world but slowly. A more significant trend, however, will be the increased use of CMOS to create better old functions as well as new IC functions.