英语外文

Much has been written about the concept of a \"system on a chip,\" the

ever-increasing integration of logic and analog functions on one silicon die or chip. This paradigm is about to change. The results of work by universities, national labs, and companies such as Motorola, Inc., are paving the way for a true system on a chip, or SOC. These new SOCs will not only analyze data, but will measure, analyze, and react to their environment.

The integration of power and analog elements with a CMOS microcontroller unit (MCU) has been possible for several years. Products have been

introduced such as an integrated 68HC05 motor controller with integral power devices in an H-bridge configuration (1990). In 1993, a product called a System Chip MCU was introduced that provided a Society of Automotive Engineers J1850 interface, including the physical layer. This chip could

withstand 40 V, based on the combination of power and analog capability with the MCU. However, the system input was not included in previous monolithic designs.

What is the most recent development that promises to truly enable a system on a chip? It is the ability to combine CMOS and MEMS

(microelectromechanical systems) structures into one process flow. Photo 1 illustrates a 68HC05 microcontroller with a 100 kPa pressure sensor integrated onto a single silicon die. A likely application is a side air bag sensor.

A pressure sensor, inside the door panel of a car, could detect the change in pressure when the panel crumples under an impact. The ability to program the onchip microcontroller will enable the auto manufacturer to embed the control algorithm inside the chip. To complete an entire system, only a mechanism for actuating the air bag need be added. This actuation capability could be yet another step in the continuous integration of silicon and

electronics/electromechanical systems. This platform provides a first step in the integration of electronics with electromechanical structures and at the same time raises several issues that must be resolved before a low-cost, high-quality product can be mass produced. One of these issues is that of testability.

Typical logic circuits have many years of accumulated test data that can be used as a foundation for building the next generation of product. With sensors,

however, very little previous technology can be reused. The reasons are the relative infancy of sensor technology and the uniqueness of each type of sensor. For example, the technology used to measure pressure (a thin diaphragm with integral strain gauge) is much different from that used for

measuring acceleration (a proof mass forming a moving capacitor). The testing technology is different as well. Pressure measurements require a pressure source to be connected to the sensor; acceleration or shock detection requires shaking the device at some known frequency and force.

System Configuration

To develop a proof-of-concept vehicle (see Figure 1), a 100 kPa pressure sensor was integrated onto Motorola's standard 8-bit 68HC05 microcontroller core along with the associated analog circuitry [1]. To this basic core was added analog circuitry for signal conditioning, a voltage and current regulator, and 10-bit A/D and 8-bit D/A converters. A temperature sensor was also incorporated into the design for compensation purposes.

The pressure transducer is temperature dependent and has an inherent nonlinearity. To increase the accuracy of the system, a calibration or conditioning algorithm must be programmed into the microcontroller.

The pressure transducer's output is conditioned by a variable gain and input offset amplifier that is controlled by the program stored in the MCU. The A/D converter is used to read the temperature sensor's and the pressure

transducer's outputs. The band gap voltage regulator supplies a constant voltage for the pressure sensor, amplifier, and A/D converter. The band gap current regulator provides a constant current source for the temperature sensor.

Calibration Method

The MCU calibrates and compensates the pressure sensor's nonlinearity and temperature drift. To provide the maximum accuracy, an A/D input resolution of 10 bits was chosen and the calculation resolution was set at 16 bits, fixed point. To calibrate span and offset and compensate the nonlinearity of the sensor output, calibration software performs a second-order polynomial correction of sensor output described as: Vout = c0 + c1Vp + c2Vp2 (1) Cp = (c0, c1, c2 ) (2)

where:

Vout = calibrated output

Vp = uncompensated pressure sensor output

To compensate the temperature dependency of Cp, calibration software is used to calculate Cp using a second-order polynomial fitting equation to temperature:

c0 = c00 + c01Vt + c02 Vt2 (3) c1 = c10 + c11Vt + c12 Vt2 (4) c2 = c20 + c21Vt + c22Vt2 (5) (6) where:

Vt = temperature sensor output

The Cts are read during the calibration procedure and stored in EPROM. The MCU calculates Cp from the temperature sensor output, Vt, and Ct. Cp is then used to calculate the calibrated pressure sensor output using the pressure transducer's output, Vp.

Calibration Procedure

The calibration system first adjusts the gain and offset of the amplifier to use the full A/D range. Then the characteristics of the uncompensated pressure sensor output are examined over several temperature points. At each

temperature, a second-order polynomial described in Equation 1 is obtained by least square fitting and the coefficient set, Cp, is determined. After

completing the calculation of Cp over all temperature points, Ct is determined by the least square fitting of Equations 3, 4, and 5 to determine Cp over the temperature points. At present, at least three separate temperature sampling points are required to complete the fitting calculation.

Figure 2. The uncompensated output of the sensor-based system on a chip is plotted at four different temperatures.

Characteristics

Figure 2 shows the uncompensated sensor output characteristics over various temperatures after adjusting gain and offset. Based on these data, the

coefficients for calibration were calculated and written into the onchip EPROM by the calibration system. The compensation value was rounded off to 8 bits. Figure 3 shows the calibrated and compensated output of the integrated MCU. Figure 4 shows the error from expected values. Since 1 bit is 0.4% error, the result indicates the error is within 0.4% of full-scale output.

Figure 3. Compensated output of the system on a chip is improved through testing and calibration at three temperatures.

Test Issues

Several issues are raised by this initial work, including the different types of testing required, unique test equipment, and the need for multipass testing. To make a low-cost integrated solution possible, these concerns must be addressed.

The integration of a physical measurement function onto the already complex mixed-mode analog-digital chip raises the need for an additional type of testing. The physical medium being tested must be applied to the device and the

response must be measured. Measuring the response to a physical stimulus is not a

Figure 4. Bit error in the compensated output is within 1 bit at both 30°C and 85°C

standard test for the semiconductor industry, especially under multiple

temperatures. Standard equipment can test the digital and analog portions of the chip, but the application of a physical stimulus and the procedure of

heating and cooling the device under test rapidly and accurately drive the need for a modified and unique tester. These testers are one of a kind and are not available as a standard. The tester therefore represents a large part of the final unit's cost.

Not only are the testers expensive, but the throughput is limited. This raises the cost of each part because of the increased depreciation costs allocated to each device. The cost is further increased by the need for multipass testing. Remember that each part is first tested, using at least three different temperatures, to determine the transducer's output characteristics over temperature. Then these values are used to derive the compensation

algorithm, which is loaded into the onchip EPROM. To complete the cycle, the device is once again tested over temperature to prove accuracy. Hence, not only is a special tester required, but it becomes a bottleneck since it must be used twice to complete each device—once to measure the characteristics and a second time to verify the result.

Future Directions

Finding ways to reduce the cost of testing is one of the keys to making a low-cost integrated sensor and MCU a reality. Ideas that could prove promising include:

Thoroughly characterizing the design Limiting the operating temperature Limiting the accuracy

Programming the MCU to take data during testing

Loading the test and compensation algorithm into the MCU before testing

Since this is a first proof-of-concept device, further characterization could

provide a way to limit the number of temperatures required for compensations. Limiting the operating temperature range could also reduce the number of temperatures required for compensation testing. Data shown in Figure 3 indicate a 5% accuracy from 5°C to 25°C. Another potential cost reduction step would be to use the MCU's programmability for data logging during test. By storing the compensation program in the onchip EPROM prior to test, and then logging the uncompensated output into the EPROM during test, it might be possible to develop an algorithm for a one-pass test over temperature.

Without a breakthrough in lowering the cost of testing this new integrated sensor and MCU, the system designer may be limited to the continued use of the present day solution—separate MCU and sensor. ----------

All the DS18B20 sensors, used for the multipoint test temperature, are

connected with MCU on one of IO bus, and temperature data are collected by turns. If the system has a large amount of sensors, the time of MCU used in processing the temperature data is obviously prolonged, so the cycle of alternate test gets longer. In this paper, a new method that DS18B20 are

rationally grouped is presented, and some measures are taken in software; as a result, the speed of alternate test advances distinctly.