Compositional neural-network modeling of complex analog circuits

Abstract

We introduce CompNN, a compositional method for the construction of a neural-network (NN) capturing the dynamic behavior of a complex analog multiple-input multiple-output (MIMO) system. CompNN first learns for each input/output pair (i, j), a small-sized nonlinear auto-regressive neural network with exogenous input (NARX) representing the transfer-function hij. The training dataset is generated by varying input i of the MIMO, only. Then, for each output j, the transfer functions hij are combined by a time-delayed neural network (TDNN) layer, fj. The training dataset for fj is generated by varying all MIMO inputs. The final output is f = (f1,. . ., fn). The NNs parameters are learned using Levenberg-Marquardt back-propagation algorithm. We apply CompNN to learn an NN abstraction of a CMOS band-gap voltage-reference circuit (BGR). First, we learn the NARX NNs corresponding to trimming, load-jump and linejump responses of the circuit. Then, we recompose the outputs by training the second layer TDNN structure. We demonstrate the performance of our learned NN in the transient simulation of the BGR by reducing the simulation-time by a factor of 17 compared to the transistor-level simulations. CompNN allows us to map particular parts of the NN to specific behavioral features of the BGR. To the best of our knowledge, CompNN is the first method to learn the NN of an analog integrated circuit (MIMO system) in a compositional fashion.

A Review of Machine Learning Techniques in Analog Integrated Circuit Design Automation

“…For example, 20,000 examples (dataset size) were needed in [13] to train the NN and predict results for a similar circuit that was designed in [14,15] using only 9000 and 16,600 different examples, respectively. On the other hand, only 1600 data points were sufficient in [12]. However, the previous numbers cannot be directly compared since the way the datasets were generated out of circuit simulators differed significantly from one study to the other.…”

“…In [18] also, the same feature was sampled d times before feeding it to the NN. A two-step approach wherein three behavioral features were considered separately for training, then recomposed into a single output in a later stage was adopted in [12].…”

“…An interesting case is the CMOS band-gap voltage reference studied in [12]. It includes an amplifier, with additional MOS transistors, diodes, and resistors.…”

“…Neural networks (NNs) of different complexities (shallow and deep) using both supervised [10][11][12][13][14][15][16][17][18] and reinforcement [19][20][21] learning techniques have been proposed in the past six years. These methods have been boosted by the progress in the development of high-performance computing hardware and are gaining attention in the scientific community thanks to their ability to model complex nonlinear problems [22]; 2.…”

Optimization techniques for analog and RF circuit designs: an overview

“…Neural networks are very effective tools used in various design steps of analog and RF circuits [43][44][45]. They are as a 'black-box' modeling and are exerted in modeling of onchip inductors [46], semiconductor devices [47], conventional analog circuit building blocks [48,49], analog IC sizing [6,50], and in crucial RF circuit blocks such as power amplifiers [51], RF front-end receivers [52], low noise amplifiers [53], voltage-controlled oscillators [54,55], and multiple-input multiple-output (MIMO) systems [56,57].…”