Micro-machined high capacity silicon load cells

The aim of the research presented in this thesis is to improve the performance of high capacity conventional load cells or force sensors by using silicon as the base material. Silicon is used because it offers the possibility of realising small, light, low cost and high performance mechanical sensors. The goal is to measure loads up to 1000 kg with an accuracy of about ±0.03 % of the full-scale output (fso). From an analysis it appears that high capacity silicon load cells need a spring element which is compressed, because otherwise the chip will break under loading. In order to eliminate for the effect of changes of the position of the loading point on the chip, the choice is made to make a sensing system which integrates the force distribution on the chip (distributed sensing). For these kind of chips a package is designed in which the influence of side forces are eliminated. Three different sensing principles are investigated: piezoresistive, capacitive and resonating. Of these principles the following aspects are discussed: sensitivity, noise level, temperature coefficient of the offset, temperature coefficient of the sensitivity, long-term stability, possibility for distributed sensing and costs. It is shown that the resonating principle is not suitable for distributed sensing. As far as costs and accuracy are concerned, it is concluded that the piezoresistive principle is a low-cost solution. The accuracies which can be achieved are medium. For very high accuracies the resonating principle is preferred. However, the production process of resonators is complex and therefore expensive. The capacitive principle forms a balanced solution in-between the piezoresistive and resonating principles. A silicon load cell with two poly-Si strain gages is realised from one wafer. One gage is compressed by the applied load. The other is used for the compensation of the temperature coefficient of resistivity, the in-plane stresses due to stretching and bending of the chip and for same changes in zero load resistor values. It is theoretically shown that the ratio of resistances of both gages is a linear function of the total force and is independent of the force distribution on the chip. A finite element model is made which is able to accurately predict the sensitivity of the sensor. Hysteresis is tested at four temperatures between 20 ºC and 50 ºC. The measurements show a hysteresis error within ±0.138 % of the fso. Creep at 1000 kg is smaller than 0.01 % of the fso. Repeatability at 1000 kg is within ±0.04 % of the fso. In addition to the chip with poly-Si gages a same type of load cell is developed the only difference being that the gages consist of monocrystalline silicon. Now, on top of the compressed gage a wafer is bonded to reduce the influence of slip between the chip and the package. From hysteresis measurements at four temperatures between 20 ºC and 50 ºC an error within ±0.058 % of the fso is obtained. Creep at 1000 kg amounts to 0.02 %-0.16 % of the fso. Repeatability at 1000 kg is within ±0.10 % of the fso. Probably, the relatively high values of the creep and repeatability errors are due to the high temperature coefficient of the sensitivity. Furthermore, a capacitive silicon load cell is realised. The chip consists of 1600 poles which carry the load. In between the poles 1600 capacitors are formed to measure the 196 Micro-machined high capacity silicon load cells deformation of the chip under loading. It can be shown that sum of the reciprocal values of these capacitors is independent of the force distribution on the chip and is a linear function of the total applied force. However, in practice it is impossible to measure 1600 capacitors individually. Therefore, these are clustered into 25 groups, each group consisting of 64 capacitors connected in parallel. It is shown that for this case the sum of the 25 reciprocal values of the capacitors is almost independent of the force distribution. This can be no obstacle for achieving the accuracy of ±0.03 % of the fso. The chip consists of two bonded wafers. On the bottom wafer, the electrodes of the capacitors are placed which are made of a titanum/platinum layer. The highly conductive top wafer forms the common electrode for all capacitors. It is shown that hillocks (spikes) in the metal are smaller than 100 nm so that for an electrode distance of 1 mm electrical short-circuits are not expected. The capacitances are determined with the help of a Modified Martin oscillator circuit. The influence of parasitic capacitances and resistances in the circuit is investigated analytically. The capacitive load cell can only be tested up to 500 kg. For higher loads short-circuits between the electrodes appear. At room temperature a hysteresis error within ±0.014 % of the fso appears. Creep is smaller than 0.01 % of the fso. The temperature dependence and repeatability cannot be tested due to stability problems with the Modified Martin oscillator. As silicon is a brittle material, a design is realised in which the force is first transformed into a fluid pressure. The axi-symmetric design consists of a steel boss that is attached to a membrane. Both parts enclose a thin fluid layer of about 2 mm. The force is applied to the centre of the boss. It is shown that the differential pressure between the fluid and surrounding air is a linear function of the force. Besides, it appears that the relation between pressure and force is independent of the elastic properties of the membrane and is only a function of the radii of the boss and the membrane. Due to this it can be expected that the transformation has a small dependence on hysteresis and creep. The results of the analytical analysis are in agreement with the results of the finite element calculations. The pressure is determined with a commercial high precision silicon pressure sensor in which resonators measure the deformation. It is shown that there is good agreement between the calculations and measurements. The hysteresis is within ±0.016 % of the fso. Creep is smaller than 0.01 % of the fso. The repeatability error is largest at 1000 kg and amounts ±0.18 % of the fso. It can be concluded that silicon load cells are a suitable alternative to standard conventional load cells, because accuracies of ±0.03 % of the fso are feasible. This conclusion is certainly true if the realised silicon load cells are compared to miniature conventional load cells with a typical accuracy of ±0.25 % of the fso