Theresa Kramer - Ph.D. Dissertation Defense, Thursday 3PM

Theresa Kramer tkramer at
Tue Apr 22 15:14:21 PDT 2003

Department of Applied Physics - University Ph.D. Dissertation Defense

Low Frequency Noise in Sub-100nm Silicon Structures

Theresa A. Kramer
Advisor: Prof. R. F. W. Pease

3:00 PM
Thursday, April 24, 2003
Center for Integrated Systems Extension Auditorium (CIS-101X)
Refreshments at 2:45 PM


Low frequency noise in bulk MOSFETs is worse than in bipolar and JFET
devices due to the effect of traps near the silicon/silicon dioxide
interface. As device dimensions scale into the nanometer regime,
deviations from constant-field scaling cause an increase in the
average trap-induced noise. Novel device configurations may allow us
to reduce this noise and can be used to explore its lower limits.

Physical separation of electrons and traps should reduce low
frequency noise in MOSFETs by reducing electron/trap interactions. We
have built depletion-mode surrounding-gate transistors that operate
with the surface in accumulation or depletion and have simulated 1/f
noise as a function of gate voltage. Simulations that include
fluctuations in electron number and mobility and only consider
electrons within one mean free path of the interface correctly
predict the experimentally observed noise. Simulations that include
all electrons or do not include both types of fluctuations do not.
Experimentally we observed not only 1/f noise, but also excess
Lorentzian noise near threshold; we attributed this to single
electron trapping. This is consistent with our observation of random
telegraph signals in the time domain.

Nanometer-scale MOSFETs should exhibit very low trap populations,
even zero in some cases, which should significantly affect the noise.
We have built cylindrical surrounding-gate transistors with 0.018
square micron channel area, which is smaller than the size in which,
on average, one trap would be active at typical trap densities. We
observed a reduction in noise by two orders of magnitude when a trap
is rendered inactive by biasing. In six of seven devices, the
measured power spectral density of the drain current is more than an
order of magnitude lower than that predicted using typical trap
densities but still has regions of Lorentzian and 1/f shape. The
near-1/f shape and presence of random telegraph signals in the drain
current indicate at least five active traps. This larger than
expected number of active traps but lower than expected power
spectral density means the devices must be populated by many traps
which have much smaller effect on drain current than predicted.

More information about the labmembers mailing list