Wireless manufacturers look to pare costs

As the profit margins for wireless products decrease, system functionality and complexity are increasing. This divergence is putting extreme pressure on wireless-system makers to reduce manufacturing costs by exploring innovative approaches to firmware programming, protocol-independent test and real-time feedback using neural networks.

In a manufacturing environment where every penny counts, such techniques can reduce test time, increase production yields, reduce the time to diagnose and repair a product, decrease time for programming firmware and reduce field failure rates. Those gains help protect the thinning margins of wireless-system makers.

A general philosophy implemented by manufacturers is to push a particular operation to the lowest possible level. The earlier an operation is performed, the easier or less costly it becomes.

Assembly, software programming and test follow similar trends. For example, manufacturers are moving away from full functional testing at the final assembly level and more toward less complicated component or subsystem-level testing. This philosophy is also followed on basic assembly steps, where RF shielding is installed at the board-level assembly, rather than the final assembly steps.

Software programming of a device-under-test's firmware is usually the most time-consuming single operation in production. As firmware becomes more sophisticated and product functionality is expanded, the amount of firmware necessary usually increases. The industry trend is to flash the firmware and to test multiple boards simultaneously.

Some companies implement preloaded firmware in memory so manufacturers can deliver memory with factory-installed firmware. The challenge with preloading memory is that firm-ware versions must be stable and the correct version must be used to prevent costly reprogramming. So the most popular trend is to flash the memory right before it is installed into a wireless product.

Several manufacturers produce equipment that loads firmware onto memory, just before the surface-mount-technology machine installs chips on a board; equipment cost and throughput rates are a concern with this type of operation. Because of that, manufacturers are leaning toward multiple-board programming rather than preloading firmware.

One of the largest capital expenditures for a production line is test equipment. Traditionally, there has been little integration of different functionality in a tester, with RF, audio, power measurements and switching all achieved by different pieces of expensive equipment. Each function, such as GPS, wireless LAN, Bluetooth, cellular service and FM audio, required a box instrument to test.

Now, the implementation of wide-bandwidth virtual instrumentation is eliminating the need for multiple pieces of standard specific equipment. Generic hardware combined with flexible software allows a single piece of equipment to test multiple functions, thereby reducing capital expenditures for specialized test equipment.

Testing the physical layer rather than the protocol layer of a wireless device is another way to shave costs. Protocol-layer testing is expensive because of false failures, increase in test time and capital expenditure for unique test equipment. Since the protocol layer is typically implemented in software, there is no need to test every device produced on a production line. Therefore, software testing is done in R&D and reliability labs rather than on a production line.

Another concern is equipment utilization. If a separate piece of expensive equipment is needed for each function, each piece may be used only 10 percent of the time required to test. One innovative method to improve equipment utilization is to test multiple systems in parallel. For example, a single wide-bandwidth instrument such as NI's PXI 5660 can simultaneously tune two cell phones.

Every test step has a defined probability of a failure. Many of them are no-fault-found failures and are a function of the test system rather than the device-under-test. So OEMs want to move toward simple test parameters that result in a simplified test infrastructure. The simpler the test equipment and interface, the less likely a failure will be a no-fault-found failure.

In addition, statistical process control helps control a production process. Statistical process control can warn an operator that the tester or production system is about to cause failures.

To have an acceptable return on investment, companies want 90 percent or better production yields. To obtain this yield rate, testers must be shut off and repaired when yields drop due to no-faults-found. The challenge is to catch these testers when intermittent failures occur.

If an operator waits until the tester's yields are 50 percent, the overall 90 percent pass rate on a line is impossible to obtain. The only way this type of yield can be obtained is to use real-time statistical process control on the production floor.

Statistical process control requires constant monitoring. Several companies have implemented novel methods for monitoring a production line's yields. For example, some OEMs use PDAs that can run LabView applications to monitor and analyze test results in real-time.

Once a wireless product is in the field, a feedback system must be in place to provide the input for the manufacturer on system failures. As the product's life span becomes shorter, companies need real-time feedback to prevent them from continuing to build products with field failures that result in expensive recalls and repairs.

One method being investigated is the use of neural networks to provide a production line with real-time feedback on field failures.

The neural networks use artificial intelligence to correlate field failures with product line settings. This type of feedback system could modify manufacturing test tolerances, adjustments and even settings such as solder oven temperature to reduce field failures.

Bill Reid is R&D senior group manager of high-frequency measurements at National Instruments Corp. (Austin, Texas).

See related chart