Hi,

I have read that NQS models are needed to accurately predict some effects, such as charge-injection, or at very high
frequencies. I have some doubts on this subject, particularly with the case of charge-injection.

The basic idea is to use an NQS model if the carrier distribution in the channel does not change fast enough so as to keep
pace with the changes in the terminal voltages (gate voltage).  There are 2 scenarios:

1. An ideal very fast edge square wave turns a MOS switch ON/OFF. The ideal approximation may be achieved by using
say a modern fast signal generator with a MOS transistor which is say 100um-200um long and the process very old (say a 2um CMOS).

2. In a real scenario, the gate of a MOS is probably being driven by another MOS device on-chip. As charge-injection is an issue in switches,
and switches are always made with min. channel length, it is unlikely that the driving edge is so fast that it can change the gate potential of the
controlled MOS switch faster than the rate at which the carrier distribution in the channel can change. This is because the controlling MOS driver itself will experience inertia in responding to its own input. Additionally, there are parasitics on-chip which further slow things down.

So, when do we really need to use NQS models? It would appear that for a typical system, the risetime of any controlling signal generated
on-chip can seldom be fast enough to force the use of such a model, unless one were driving a device which is considerably slower (very large L as in 1.).

Regards
Vivek

As I understand it, the problem with charge injection is that QS mos models have no idea what to do with the channel charge when the gate is shut off.  For example, in BSIM3/4, the XPART parameter is used to allocate the charge to the drain or source in the ratio 0/100, 40/60, or 50/50.  When you turn a BSIM transistor off, the charge flows out the respective terminal -- regardless of the impedance connected to that terminal!