Figure 3: Principle of the narrowband infrared detector on organic semiconductors.

© Senorics

Conventional spectrometers that work in the near infrared are based on a grating monochromator in combination with a line sensor made of indium gallium arsenide. Both are very expensive to manufacture; the prices of such devices can run into the five-digit euro range. However, the high resolution of a few nm achieved by these devices is not absolutely necessary for alcohol and sugar measurement. Here the requirements can also be lower.

Dr. Ronny Timmreck and his team have found a practicable solution for an extremely cost-effective sensor at Senorics GmbH in Dresden. The company is a spin-off from the Institute of Applied Photophysics (IAPP) at the Technical University there. The invention does not require expensive InGaAs detectors and gratings, but uses spectrally selective detectors made of organic semiconductors - a key area of research at the IAPP.

Figure 4: Practical version of the alcohol and sugar meter.

© Senorics

On a 20 x 20 mm² substrate, 16 individual elements are placed, all made of the same material but with different thicknesses - just enough to fit vertically half a wavelength of a very specific infrared line(Fig. 3). In this way, such a structure acts as a wavelength-selective IR detector; the achievable half-wavelength widths of 20 to 30 nm are narrow enough for the intended purpose. The individual elements are dimensioned in such a way that they just detect the IR spectral lines relevant for alcohol and sugar.

In order to achieve accuracy in the sub-percent range, the evaluation of a few wavelengths is sufficient for most applications - a maximum of 20, typically 4 to 10. Only very small sample quantities are required. The signals of the individual elements are digitized with 16-bit resolution. A microprocessor applies the various chemometric models to the measured values and uses them to determine the sugar and ethanol content. The final result is a handy, easy-to-use device at a price of less than 300 euros(Fig. 4), affordable for even the smallest brewery.

STMicroelectronics: Computing core for machine learning

Inertial sensors in MEMS technology, which record accelerations and rotations in three coordinate directions, are already widely used. They are often found in battery-powered devices - smartphones, wearables, etc. - where they monitor body movements, for example, and then transmit the measurement data somewhere.

The raw data recorded in the process can quickly grow to considerable volumes. Sending it directly would require a lot of energy because the transmission stage in the device is the largest power consumer. The battery would quickly run out. Prior evaluation and concentration could save a lot here, but would require a lot of computing power in the processor, which would also require a lot of energy. So not much would be gained.

STMicroelectronics has therefore developed a fundamentally different approach in the 'LSM6DSOX' component. A normal microprocessor is not ideally suited for the special calculations to be carried out here. Finite state machines (FSM) work much more efficiently, with up to 16 of them integrated on the same chip, all independently programmable and designed for deductive algorithms. A machine learning core (MLC), optimized for inductive algorithms, is also integrated. Both can also be coupled together. Figure 5 shows the block structure of the module.

Figure 5: Block structure of the LSM6DSOX inertial sensor.

© Senorics

These elements manage with much less operating current. A rough estimate: At a data acquisition rate of the inertial sensor of 100 Hz, the 16 FSMs and the MLC together only consume around 10 to 15 µA. A normal processor with an ARM M4 core and 100 MHz clock rate would require around 20 mA for the same tasks, more than a thousand times as much.

The calculations in question are mainly pattern recognition algorithms. What physical activities does the user perform: resting, walking, running, cycling, riding, weight lifting, etc.? The sensor module compares the registered movement sequences with previously stored characteristic patterns and then recognizes what the wearer of the device is currently doing. It can then decide - on the basis of previously entered threshold values - whether or not it should transmit the recorded data. For example, it could differentiate between normal sport and a fall.

Figure 6: Size only 3 x 2.5 mm: LSM6DSOX inertial sensor.

© STMicroelectronics

Not only competitive athletes will be among the users. There are also countless potential applications in industry. For example, on an assembly robot, where the number of movements or processed parts can be determined as information on wear and tear and for predictive maintenance.

As the power-hungry microprocessor now has much less to do, it can remain in sleep mode most of the time and is only woken up when special tasks have to be carried out that exceed the capabilities of FSM and MLC. This saves a lot of power. The battery or the entire device can become smaller, or the operating time per charge can be extended. Other features: Up to four external sensors for other variables can be connected to the module. Data is transmitted via an I³C interface. The size is 3 x 2.5 x 0.8 mm³(Fig. 6). The current consumption is only 0.55 mA.

Trinamix: Fiber optic sensor for absolute distance measurement

Distances have to be measured in countless industrial applications. There are already a variety of measurement methods available today, but there are always cases where the established methods are not suitable and special solutions are required - for example in environments with high temperatures or strong electrical or magnetic fields. This is where fiber optic sensors are advantageous. With these, no currents flow at the actual measuring point, but only in the optical transmitters and receivers at a greater distance.

There are two different types: With "intrinsic" ones, the fibre itself reacts to the measured variable; with "extrinsic" ones, it only serves as a feed line to the sensor element; a physical effect takes place in the latter, in which the measured variable influences the light in some way - for example in amplitude, phase, polarization direction or spectrum, after which it flows back to the evaluation station either via the same or a different fibre.

The 'XperYen' distance sensor developed by Dr. Celal Mohan Ögün and his team at Trinamix GmbH in Ludwigshafen belongs to the extrinsic class. It uses a new measuring principle called beam profile analysis, invented at BASF. After this concept proved to be extremely promising, the division was spun off as a separate subsidiary. In contrast to the common triangulation method, the optical transmitter and receiver are located in the same housing. Only a very small measuring head is located at the actual measuring point, which only contains a mechanical holder for the fiber ends and a simple imaging lens. It has a monocular design, i.e. the single lens is used both to collimate the transmitted light beam and to collect the light reflected from the object being measured.

Figure 7: Principle of beam profile analysis.

© Trinamix

In the evaluation unit, a light source emits visible or infrared light into a central transmission fiber that leads to the measuring head, where it is bundled by the lens and falls onto the object to be detected. The very small spot of light is imaged onto the surface of an optical sensor using the same lens - deliberately not in the focal plane, but deliberately blurred so that it expands into a larger circle.

The light then enters the ends of several receiving fibers around the center in the same optical cable(Fig. 7), runs through them to the evaluation unit(Fig. 8) and falls on a suitable optical detector. In principle, this could be a normal CMOS image sensor; however, such a sensor would be too slow for many fast industrial processes and would also require complex software for analysis. Several single-cell photodiodes in a sophisticated arrangement are more favorable; they react significantly faster and manage with simple analog electronics.

Figure 8: The compact evaluation unit of the XperYenZ distance sensor.

© Trinamix

The key feature of beam profile analysis is that it works completely independently of the diameter of the light spot. This means that both a diverging LED and a collimated laser can be used as a light source. The starting point for calculating the distance is the relative intensity distribution between the individual measurement paths. The absolute light intensity and the color or reflectivity of the object do not play a role, nor do the properties of the fibers such as type, length or material. It is therefore possible to use inexpensive plastic fibers and temperature-stable glass fibers, as well as single or multi-mode fibers of any length. Moving mechanical parts such as adjustable lenses or iris diaphragms are not required. The measuring range is 25 to 120 mm, the linearity error remains below 1 mm.

The potential applications of XperYenZ are primarily in the field of highly automated production, where concepts such as Industry 4.0 are to be implemented. In the packaging industry, a typical example is checking whether there is a product in every cookie or tablet pack. The savings potential of a closely monitored and fully automated production process exceeds the costs of XperYenZ fiber sensors many times over.