Man vs. machine: many PCB designers have a love-hate relationship with their design tools: they love the time automation saves them, but hate giving up control. An EDA vendor looks at the delicate balance between too much automation and not enough.

Someone once said that in product design, having the idea is the hardest part. But he never had to lay out a 32-bit processor board by hand using tape and tracing paper. Actually, the hardest part of product design is balancing the complexity of the design with the time needed to implement it.

PCB design is an area that stems from simple fundamentals--connecting a series of points in the right order to complete a defined circuit--but the difficulty of the process relates fundamentally to the size of the task. The more connections to be made, the more complex the task of designing the PCB and the more time it takes to iterate through the process.

It's no accident that PCB design was one of the first electronics design tasks to which computer technology was applied in an effort to automate the process. EDA had its roots in the need to handle the explosion in circuit complexity brought about by the advent of digital electronics and, in particular, the rise of the microprocessor in electronics design. Traditional manual design techniques became too cumbersome to manage when faced with the onslaught of higher pin counts, multiline signal buses and increasing numbers of board layers.

Today, board-level EDA is a mature industry and EDA tools are ubiquitous in the design of electronic products. In fact, it's unthinkable that any modern electronic design would be done without the use of computer-based tools. However, automation in the design process is more than feeding data to a software algorithm. It must assist the designer to get a task done more quickly and efficiently, without hindering the desired outcome by assuming too much control.

The Quickest Route

Most PCB designers today take the inclusion of an autorouter in a PCB design system as a given. An autorouter's job is to take the net-level connectivity of the circuit and convert it into finished routes that comply with the design constraints set for the board. Routing is an ideal candidate for automation because it is an iterative process. When large numbers of nets are involved, the time taken to iterate through each connection and find a suitable path can be considerable, so the benefits of automation in this area can be great.

Early autorouters were notoriously difficult to set up, requiring an in-depth knowledge of the application in order to achieve acceptable results. While they had the potential to remove the complexity associated with routing large numbers of tracks, they introduced a new layer of complexity to the design process--that of understanding how to set up the constraints necessary to direct the autorouting process itself.

From the PCB designer's perspective, the time saved by the automation process was often lost in to the time taken to set up the autorouter; the complexity was simply moved from one area to another.

EDA vendors responded to this by providing tools to aid the setup process, and to introduce technology to automatically analyze the board and configure the autorouter appropriately. While this had the effect of making the autorouter easier to use, it meant that some control over the process was taken from the PCB designer and given to the EDA tools vendor.

In order to make the automation useful, by necessity the tool vendor must hide complexity from the user and decide the appropriate way to handle certain situations. Striking a balance between lowering complexity and providing enough control over the process is a perennial problem facing all EDA tool developers. In autorouters, this balance has largely been achieved by a combination of better configuration tools and a greater understanding of the technology by both the vendors and the users.

Contrasting with autorouting is the area of automatic component placement. Like routing, placement is an iterative process involving many conflicting constraints, but to date, automatic placement tools have not achieved widespread acceptance in the industry. While there are a number of autoplacers in the market, and many integrated within PCB design tools, they are not yet perceived as an integral part of the board design process.

This is in part because the results achieved by autoplacers are in many cases not worth the effort of invoking the automation. But this is not necessarily, as is commonly believed, a product of the quality of the tool or the sophistication of the algorithms used.

Autoplacement differs from autorouting in the constraints that drive the process. With routing, the constraints are largely dictated by the characteristics and parameters of the board itself. These properties are easily modeled within the PCB design environment because they are properties of the board design--the number of layers, spacing between tracks, connection distance, layer direction, etc.

The constraints dictating component placement, however, can be driven by mechanical considerations such as the shape of the product casing, by ergonomic issues such as the positioning of buttons and connectors, or by other concerns such as heat dissipation and optimization for pick-and-place machines during fabrication.

The fact that these constraints are properties of the wider product design, not of the board design, means that trying to accurately model these properties in the PCB space to drive placement introduces significant overheads for the PCB designer. Until electrical and mechanical CAD systems are fully integrated, and the design constraints imposed by both realms can be exposed in an integrated system, the application of automation to board-level component placement is an inefficient prospect.

Too Much of a Good Thing?

The art of design automation is, in effect, choosing the right aspect of the design process to automate, and what level of complexity to expose within the automated process. When a process is automated, the input of the designer is redirected away from hands-on implementation and into creating the constraints that define and drive that implementation. Instead of formulating a strategy in his head and then using the tools to apply that strategy, the designer creates and formalizes the strategy within the design tools and then the automated processes do the work of implementation.

To succeed, a design environment must support the definition of all the parameters needed to model the process to be automated. It must then be able to translate the design constraints into actions that produce the result the designer intended. If it's too difficult to set up the appropriate constraints, or the results aren't what was intended and require significant manual rework, then the automation can actually hinder the design process.

One of the hardest things from an EDA tool vendor's perspective is understanding which parts of the process should be fixed and which should be controllable by the designer. The challenge is to balance user control with productivity.

One strategy is to provide a layered approach to the automation. At the top layer the system automatically configures itself to use appropriate defaults for all design parameters--the so-called "click and go" operation. Beneath this layer, basic configuration options can be exposed to control the operation at a coarse level. Lower layers can then progressively give finer control over the automation process.

While there is a tendency for designers to want this layering to go right down to very fine levels of control, the law of diminishing returns dictates that a sensible cutoff must be imposed. Every exposed parameter increases the complexity of the coding required in the software. Also, the interdependence of many parameters means that changing one often has undesirable effects on others, leading to unpredictable results. In the end the designer must be willing to relinquish some level of control and trust the automation to do its job.

Ultimately, picking the right balance is an iterative process. EDA vendors progressively gain a better understanding of the design process they are trying to automate, and designers become increasingly more familiar with the automation process and the benefits it brings.

One criticism often leveled at the EDA industry is the lag between the development of new electronics technologies and the release of the automation tools needed to support it. This has been particularly noticeable in chip-level design where the wide-scale adoption of new manufacturing processes is often delayed by the lack of suitable tools. Compared with chip design, changes in board-level technologies are often not as radical or as rapid. Nevertheless, PCB tool developers face similar dilemmas when addressing the need for increased automation in the design process to support new device and manufacturing technologies, and reduce time to market.

The problem resembles the "chicken or the egg" paradox. Do EDA companies invest money and R&D effort trying to anticipate future trends and have tools ready when these trends become manifest? Or do they spend their resources improving existing tools and features and react to changes as they occur? They must do both, or at least achieve the best balance they can between the two.

The reactive approach is a far easier path for EDA companies, as it allows them to address the current problems faced by their customers. Solving a new problem and providing automation to speed existing design bottlenecks has immediate benefits to the mainstream of designers.

But the only constant in the electronics industry is change, and the pace of change continues to pick up. Time-to-market pressures are forcing companies to examine every aspect of the design chain to squeeze development cycles. Resting solely on reactive EDA tool development is an enormously risky approach. When new technologies and methodologies come along that offer significant competitive advantages, companies reassess their tool needs. If the entrenched EDA vendor can't satisfy these needs quickly enough, they stand to lose customers.

The success of any new design technology or methodology is never assured. However the early availability of good tools that allow engineers to take advantage of new devices or manufacturing technologies can increase usage. By being proactive, EDA vendors can contribute to the successful uptake of the very technology they are gambling on.

When you consider that we've come from circuits containing just several transistors to several hundred million in the space of less than half a century, it almost defies belief that this complexity can be managed and harnessed by an engineer working in an office on a desktop computer.

Increased automation of the design process in and of itself does not necessarily drive innovation in the electronics industry. What it does is raise the level of abstraction that a designer works at, empowering them to do what they do best. Design time can then be spent adding value at creative level, rather than dealing with the complexities of implementation.

EDA tool vendors must strike a balance between making it easy to set up the constraints for design automation and allowing the designer enough control to influence the outcome to suit their needs. Designers must be willing to trade some level of control for the benefits that increased automation brings to the design process.

The ultimate goal of all design automation is to maximize the effectiveness of the most valuable of all design assets--the creativity of the designer.

ROB IRWIN is manager, brand strategy, at Altium Limited. He has a bachelor's of engineering (electrical) from the University of Sydney, Australia, and over 20 years experience in the electronic design industry. Irwin can be reached at [e-mail].