Abstract
An integrated machine-learning based adaptive circuit for sensor calibration implemented in standard 0.18μm CMOS technology with 1.8V power supply is presented in this paper. In addition to linearizing the device response, the proposed system is also capable to correct offset and gain errors. The building blocks conforming the adaptive system are designed and experimentally characterized to generate numerical high-level models which are used to verify the proper performance of each analog block within a defined multilayer perceptron architecture. The network weights, obtained from the learning phase, are stored in a microcontroller EEPROM memory, and then loaded into each of the registers of the proposed integrated prototype. In order to verify the proposed system performance, the non-linear characteristic of a thermistor is compensated as an application example, achieving a relative error er below 3% within an input span of 130°C, which is almost 6 times less than the uncorrected response. The power consumption of the whole system is 1.4mW and it has an active area of 0.86mm <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> . The digital programmability of the network weights provides flexibility when a sensor change is required.
Highlights
In sensor production it is desired that all the sensors have the same well-defined characteristic with a certain accuracy
This paper presents an integrated CMOS mixed-mode machine learning model based on a multilayer perceptron (MLP) configuration, providing an efficient and robust method to compensate any kind of non-linear output sensor response
A mixed-mode machine learning based conditioning circuit integrated in standard 0.18μm CMOS technology has been presented in this paper
Summary
In sensor production it is desired that all the sensors have the same well-defined characteristic with a certain accuracy. The calibration can be performed in the conversion step using non-linear ADC converters [7]–[9], so that both, the digital conversion and linearization are performed simultaneously These techniques minimize digital implementation costs, but have no flexibility when using different sensors. An optimal solution is a mixed-mode integrated ML model with analog processor units, to minimize power consumption and area, and digital programmability, to facilitate the reprogramming of the model parameters This approach constitutes a flexible solution at a low cost in terms of area, power consumption and computational complexity, being a valuable choice for adaptive sensor processing in embedded applications [14]–[16].
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have