Symmetry Constraint based on Mismatch Analysis for Analog Layout in SOI Technology
The conventional tools for mismatch elimination such as geometric symmetry and common centroid technology can only eliminate systematic mismatch, but can do little to reduce random mismatch and thermal-induced mismatch. As the development of VLSI technology, the random mismatch is becoming more and more serious. And in the context of Silicon on Insulator (SOI), the self-heating effect leads to unbearable thermal-induced mismatch. Therefore, in this paper, we first propose a new model which can estimate the combination effect of both random mismatch and thermal-induced mismatch by mismatch analysis and SPICE simulation. And in order to meet the different sensitivities of different symmetry pairs, an automatic classification tool and a configurable optimization process are also introduced. All of these are embedded in the floorplanning process. The final experimental results prove the effectiveness of our method.
The variations in the parameters of two equally designed devices can be defined as mismatch. And it is seen as a limiting factor for the function and performance of analog circuits. In general, mismatch can be classified into two categories according to the different generating mechanisms: fabricated-related mismatch and ambient-related mismatch.
For the fabricated-related mismatch, there are two more
detailed categories: (1) systematic mismatch; (2) random
mismatch. Manufacturing variations results in device
parameter variations from batch to batch, wafer to wafer
and device to device. Batch-to-batch and wafer-to-wafer
variations are common to all devices in the circuit. So,
they are regarded as the systematic mismatch. And their
effect on the circuit performance can be largely eliminated
by layout techniques such as symmetry constraint.
Therefore, the conventional tools for symmetry constraint
[1]-[4] mainly aim at automatically placing the equally
designed devices symmetrically with respect one or
several common axes in order to eliminate the systematic
mismatch. And we define these tools as geometric
symmetry tools. Unfortunately, the existence of random
mismatch weakens the ability of geometric symmetry tools
in eliminating mismatch. Device-to-device variations
which are defined as the random mismatch are caused by
some unpredictable processes during design and
fabrication phases. And random mismatch will greatly
increase as the size of transistors decreases [12]-[14].
Nowadays, the size of transistor is in the magnitude of
deep-submicron. As a result, the random mismatch will
become more severe. This brings a great challenge for
geometric symmetry tools to guarantee the matching
constraint for the symmetrical devices.
The ambient-related mismatch can be expressed as the
mismatch caused by the ambient factors like thermal
condition and noise distribution. Thermal-induced
mismatch [6]-[8] is an important one in this kind of mismatch. Because of the relatively better thermal conductivity of the silicon [8], the thermal-induced mismatch is often neglected and the analog circuits are assumed as isothermal. However, in the context of SOI, the conductivity of isolating buried oxide is often over two orders of magnitude worse than silicon [9]. Thus, even with the relatively moderate power levels encountered in typical signal path transistors, increases in the channel temperature of tens of degrees due to self-heating effect can be observed [10]. And some of the heat generated by the distinct devices will flow laterally before reaching the substrate. So, the temperatures of these neighboring transistors will also rise. The temperature gradients resulting from self-heating and thermal coupling lead to nonisothermal conditions. Literature [11] is proposed to reduce the temperature gradient on symmetrical devices. However, its optimization objective is the Minimum Average Temperature Gradient (MATG) of all symmetry pairs. This method only aims at temperature optimization and ignores the different mismatch sensitivities of devices for different use and this may still lead to mismatch on some sensitive symmetry pairs