Catalin Andrei DOBRIN

Abstract: The design of circuits in microelectronic technologies is characterized by the need to apprehend the notion of variability. Components available in integrated technologies suffer from tolerances, especially when comparing devices on different wafers. These variations in component parameters inevitably result in circuit performance variations.

While variability in performance has traditionally been managed by incorporating margins, this approach is no longer acceptable due to its wastefulness (in terms of energy and silicon area) and consequently, a lack of product competitiveness. The usual approach is to provide adjustable circuit parameters (such as the bias current of a transistor) to position each circuit precisely at the specification level.

Methods for adjusting circuits fall into two types: either a posteriori (measuring performance and finding the adjustment that yields the desired performance) or a priori (for example, adjusting the cutoff frequency of a filter by varying the capacitance to achieve the desired RxC product).

The first approach, which appears to be the most comprehensive, suffers from several limitations. It is often invasive (requiring performance measurements) and resource-intensive (requiring additional receivers to measure the level of unwanted frequencies). Furthermore, it is limited to in-situ measurable performances (for example, it may not measure NF), and it is challenging to apply when a compromise between multiple performances is involved (such as linearity and noise).

The second approach is limited to performances that can be correlated with measurable parameters. This correlation can be analytical (such as the cutoff frequency of a filter) or, for more complex metrics extracted from simulations or measurements, some works use artificial intelligence techniques to highlight these correlations.

The objective of this thesis is to advance the analytical approach to complex performances. The starting point is the formulation of the circuit equation, which allows for the equation of the performances of interest based on component parameters. The work will involve finding a way to extract, in-situ, the parameters of the components and implementing calibration levers.

It is noteworthy that these calibration-oriented works will also impact industrial test methods (extracting parameters rather than measuring complex performances) or auto-testing (Built-In Self-Test). To demonstrate the effectiveness and relevance of the proposed approach, the thesis work will conclude with measurements of a demonstrator in an advanced CMOS technology.