September 2002
The System on a
Chip
The tug of war between the demand for custom chip development and the constraints of many low-volume designs makes it your business to know what innovative system-on-a-chip designers are doing and the options this technology offers.
John S. Rinaldi, Real Time Automation
Your company plans to launch a new line of ultrasonic sensors. The design calls for significant improvements over a previous design, which is based on an 8051 microprocessor and discrete components. You have a total of four models to develop: externally programmable, discrete output; externally programmable RS-232 output; fixed internal program, discrete output; fixed internal program, RS-232 output.
Traditionally, this would require four new designs using discrete components, with four individual bills of materials, schematics, and PCB layouts. If you were going to sell 1 million units of each, you might be able to justify the engineering and tooling cost for each of the four designs. But what if the anticipated sales volume is only 1000 pieces per year for each unit? Suddenly the success of the entire product line depends on your ability to modify a single design and capitalize on every possible design efficiency.
The issues here are universal throughout the electronics industry, and not just with low-volume products. The ever-decreasing cost of computers and the demand for more power means that more and more functions are taken off the PCB and put into the Pentium processor (e.g., power management, disk controllers, and cache). ICs that were previously external are now internal.
You need a similar advantage in your ultrasonic sensor design. If there were a microprocessor that had a serial port, digital I/O pins, and an external programming interface (e.g., an I2C or USB port), you could use the processor for all four designs—and probably the same PCB as well. The only difference might be the external connections that you make to the board (read: money saved on manufacturing, engineering, and tooling).
Remember your old 8051 code? If the new processor were available with 8051 support, you could reuse and update your existing code. So the economies of software development are similarly improved.
The good news is that you really can do this. The system on a chip (SoC) business is a segment of the semiconductor industry that solves these kinds of problems.
Roll Your Own?
In the past, high-volume manufacturers would design application-specific integrated circuits (ASICs) for themselves. Of course, some people saw an opportunity to develop quasi-custom ASIC solutions, buying intellectual property from chip developers, mixing/matching components, and offering them for sale to developers as off-the-shelf products.
Only a short time ago, building a completely customized single-chip solution for an embedded application was a risky and expensive process. If there were a missed feature or a misjudgment of market requirements, it was more than just a headache. More than likely, the next project for the product development team would be new resumes.
But what if you could customize a chip solution for yourself, in low volume?
Configurable SoCs
First, consider field programmable gate arrays (FPGAs) and programmable logic devices (PLDs), which are at the extreme end of configurable processors. While the nuances of these two technologies are beyond the scope of this article, pure FPGAs and PLDs are essentially blank slates. You can burn in custom logic and effectively simulate the operation of any active device you wish to create. These devices are used for implementing customized low-volume chip solutions, in somewhat the same way you might burn software on a CD-ROM and sell it in low volume instead of pressing CDs mass production style.
Many SoC products combine multiple peripherals with a certain amount of blank-slate configurability, usually in the form of libraries predefined by the manufacturer. This allows a single piece of hardware to be programmed with a diverse array of turnkey feature options, speeding development and allowing extensive customization.
Advantage over ASICs
SoC’s configurability is a huge benefit. The process of designing ASICs always required an incredible amount of foresight into the customer’s requirements and the target market. Re-spinning silicon after rollout meant lost market share and financial doom.
And the process has not improved. As chip densities have soared, ASIC development has become even more cumbersome and expensive and is a greater gamble for the developer. Only the largest companies can afford the time, development costs, and risk for anything but the highest-volume applications.
Silicon Intellectual Property
As custom chip solutions have evolved, a market for chip dies has also grown. SoC and FPGA technologies have in turn given ASIC developers the ability to roll out designs faster and in lower volumes. So if there’s a discrete component you’d like included in your ASIC, you can probably buy the intellectual property on a royalty basis. This has increased the options available to SoC and FPGA customers.
Lucky for product developers, configurable processor architectures now provide the flexibility, performance, low cost, fast time to market, and product differentiation needed by embedded solutions. Several types of solutions are available, including soft instruction, configurable, and combo processors (which implement aspects of both soft instruction processors and configurable processors).
Soft Instruction Processors
Soft instruction processors let you customize CPU architectures by specifying instructions supported, peripherals available, and number of registers. Some vendors provide mechanisms to add, delete, and create highly tailored instructions. Design packages for these architectures sometimes include performance tools with instant feedback on the performance, die size, and power requirements of a design. With the final architecture residing in silicon, these types of architectures are well suited for high-volume, low-cost applications, which formerly would have used ASICs.
Configurable Processors
Configurable processors, also known as SoC processor architectures, are based on FPGAs. With these, you can add, modify, and extend standard and customer-derived logic engines as needed. By moving discrete logic functionality to internal FPGAs, you get a highly flexible logic solver based around a standard processor core.
With FPGA logic instead of foundry logic, you can easily revise the logic at any point in the design cycle. You can quickly create specials or custom logic configurations for potential customers, driving the batch size down to single-unit quantities.
Real-World Examples
During the height of the fieldbus wars, Lantronix’s subsidiary Synergetic Micro Systems saw the writing on the wall: The cost of embedded networking capability (especially fieldbuses) was too high in relation to the cost of the basic product. An additional communications daughterboard consumed too much cost and space to be an option for a small $250 temperature controller. The situation was further complicated by the many choices of networks (e.g., Profibus, DeviceNet, and Ethernet).
The industrial and manufacturing networking market is dominated by a few customers (e.g., GM, Chrysler, and Ford) and a small number of vendors (e.g., Rockwell Automation/ Allen Bradley, GE Fanuc, Groupe Schneider, and Siemens). Each vendor promotes different industrial networking technology, which is accepted to various degrees by the major automotive customers. Until now, offering product to customers of diverse vendors was a nightmare of conflicting physical- and application-layer standards.
SoC developers at Synergetic made two predictions based on what’s been seen many times in the PC world:
- Networking will inevitably become standard fare, instead of an option, thereby driving volumes up.
- Multiple networking options are expensive in discrete components, but not on silicon. You can place multiple network protocols on the same chip without a proportional increase in cost.
To this end, Lantronix created the DSTni (see Figure 1), a chip that combines a 186 processor with RAM, ROM, dual port memory, and several communications options (e.g., Ethernet, CAN, Profibus, and serial ports).
Figure 1. Lantronix' DSTni (Destiny) is a system on a chip that combines an x86 processor with Ethernet, CAN, Profibus, serial ports, RAM, and dual port memory, reducing the parts count and design time of multinetwork embedded devices.
|
Manufacturers of $250 temperature controllers would be able to use one chip for both processing and communications, replacing four to eight discrete components.
As with most SoC designs, the value here is realized through simplification of software development and reduced bill of materials, circuit board space, power consumption, and heat. The product comes with software libraries and tools to speed development. It’s easy to justify a communications SoC when your product line must support two or more networks.
Configurable System Logic and an 8-Bit Processor
For us grizzled veteran designers who love our 8-bit technology, Triscend developed a souped-up 8032 MCU platform in its E5 family. The E5 marries standard 8051 MCU technology to configurable system logic (CSL), which can be thought of as a matrix of gate arrays.
By combining the CSL with a library of soft modules—which include serial communications, logic functions, and display drivers—Triscend delivers an easy-to-use, customizable solution for configurable SoC applications. It brings simple 8-bit architecture together with sophisticated communications, three 16-bit timer/counters, a watchdog timer, a full-duplex UART, and three external and 10 internal interrupts.
Simple Software Configuration
Implementation of CSL is hidden from the designer. Using Triscend’s point-and-click Fast Chip Design Studio, logic modules can be dragged onto the chip as needed.
After you select modules from the Fast Chip Library (see Figure 2), Fast Chip provides you with interface code files that you include in your project.
Figure 2. Triscend's E5 combines an enhanced turbo 8032 microcontroller and CSL. The manufacturer supplies a software package that creates and downloads custom configurations that allow the chip to function exactly as desired. (Figure courtesy of Triscend Corp.)
|
These files connect the selected peripherals to the application code. No need to spend hours managing cell addresses or peripheral implementation details while fighting tight development deadlines.
Why Pick an SoC?
Any of the following reasons could be justification for considering an SoC for your next product:
- Size constraints that dictate reduced parts count
- Conflicting priorities in hardware support for multiple related products
- Development budget and schedules that permit only a partial rollout of your product line
- Desire to use existing programming code in a more sophisticated design
- High cost of materials that makes a complex design unmarketable
Editor’s Note
John Rinaldi’s white paper, “5 Issues to Consider Before Designing With a System-on-Chip,” is available free on request. Simply e-mail soc@rtaautomation.com or call 414-453-5100 with your street address. Also ask about his e-book, “Industrial Ethernet in 90 Days or Less: A Plan for Product Developers.”
SIDEBAR:
Questions You Should Ask When Considering System-on-a-Chip Devices
|
|
Perry S. Marshall, Perry S. Marshall & Associates
Selecting an architecture, tool set, and methodology from the wide range of configurable processor solutions is a huge technical and production challenge. Let’s look at some of the more important questions that the product design team should ask.
Where are we on the volume/flexibility curve?
A high-volume application usually requires specific performance and cost targets. Flexibility is a minor consideration, if it is considered at all. These types of applications will often be most effectively handled with ASIC technology generated from analysis and development tools offered by soft-instruction-processor vendors. These tools generate highly tailored, highly specialized, mass-producible dies using the set of instructions and components selected by the designer.
Lower-volume applications usually require greater flexibility, are less cost sensitive, and can tolerate a wider performance window. These types of applications are much more appropriate for configurable processors. With the ability to quickly and easily modify hardware logic right up until production, the designs can meet the needs of changing markets.
How critical is performance, and what kind of performance is important?
There’s a great difference in the performance attained by processor algorithms, FPGA logic engines, and discrete logic. If you have critical custom algorithms, they’ll run more slowly as software algorithms than as a long FPGA logic pipeline in a configurable processor.
If resolution is an issue, discrete logic can be resolved much faster than FPGA. In some cases, the customizable instruction set of a soft instruction processor can yield a 10 × performance increase over other implementations. You have to understand and benchmark the performance requirements of your application before selecting a SoC solution with the proper performance.
Is DSP a requirement?
Many more applications now require the integration of DSP technology with standard processor functions. If this is required by your application, soft instruction processor architectures typically have DSP functions that can be optionally integrated into the processor core.
Does your design require readily available standard components?
If your application uses a number of common peripherals (e.g., UARTs, timers, and interrupt logic) or specialized components (e.g., LCD interfaces), a configurable processor can not only accelerate your product development but reduce your PCB real estate, component count, and one-time nonrecurring engineering.
How important and sophisticated are the vendors’ troubleshooting capabilities?
If you’re designing a highly reusable product, the debug capabilities become extremely important. Here are some questions you should ask:
- How easily can you connect to the system?
- What kinds of specialized software are required?
- Since you won’t be using it on an everyday basis, can a novice get up to speed quickly and troubleshoot the design without extensive training?
If your product is on the high-volume/low-flexibility curve, the vendors’ troubleshooting capabilities are much less important.
Perry S. Marshall is the author of the book Industrial Ethernet: A Pocket Guide and is President of Perry S. Marshall & Associates, a consulting firm that provides support to high-tech OEMs on product definition, marketing, and lead generation. Technical articles and business tools are available at www.perrymarshall.com. You can reach him via e-mail from his Web site or at 708-788-4461.
|
|
John S. Rinaldi is President, Real Time Automation, 2825 N. Mayfair Rd., Ste. 11, Wauwatosa, WI 53222; 414-453-5100, fax 414-453-5125, jsr@rtaautomation.com.
|
|
 SUBSCRIBE NOW!
|
|
|
