Performance-Driven Unit-Capacitor Placement of Successive-Approximation-Register ADCs

Abstract

The performance of many switched-capacitor analog integrated circuits, such as analog-to-digital converters (ADCs) and sample and hold circuits, is directly related to their accurate capacitance ratios. In general, capacitor mismatch can result from two sources of errors: random mismatch and systematic mismatch. Paralleling unit capacitance (UC) with a common-centroid structure can alleviate the random mismatch errors. The complexity of generating an optimal solution to the UC placement problem is extremely high, let alone if both placement and routing problems are to be optimized simultaneously. This article evaluates the performance of the UC placement generated in an existing work and proposes an alternative UC placement to achieve optimal ratio mismatch M and better linearity performance of SAR ADC design. Results show that the proposed UC placement achieves a ratio mismatch of M = 0.695, the effective number of bits ENOB = 8.314 bits, and the integral nonlinearity INL = 0.816 LSB (least significant bits) for a 9-bit SAR ADC design.

PACES: A Partition-Centering-Based Symmetry Placement for Binary-Weighted Unit Capacitor Arrays

“…Moreover, to alleviate the second-order lithographic errors and the proximity effects, the capacitor array should be surrounded by the dummy unit capacitors [4,16]. This implies that, with this routing method and layout implementation, we can focus only on placement of unit capacitors [15,22]. …”

“…In fact, the success of resulting in better ratio mismatch was attributed to place both C0 and C1 at the central entries of the array [22]. Moreover, the placement with large overall correlation coefficient L cannot result in better performance [22].…”

“…In fact, the success of resulting in better ratio mismatch was attributed to place both C0 and C1 at the central entries of the array [22]. Moreover, the placement with large overall correlation coefficient L cannot result in better performance [22]. In addition to evaluate the placement with ratio mismatch M and overall correlation coefficient L, the linearity performance also can be used to evaluate the generated placement.…”

“…However, it had been shown that the placement with larger overall correlation coefficient L is not a sufficient criteria [21,22] and thus the Monte-Carlo simulation [23] is conducted to evaluate the random mismatch for simulating the statistical fluctuation.…”

“…To reduce the placement generation time while maintaining a low ratio mismatch M and high overall correlation coefficient L, the diagonal weighting method [16] was presented. However, several counterexamples have shown that a placement with a low ratio mismatch M and high overall correlation coefficient L may not exhibit high performance [21,22].…”