According to ITRS 2.0 2015, memory technologies will continue to
drive pitch scaling and highest transistor count. As DRAM products are
expected to reach their scaling limits by 2024, and unless some major
breakthrough occurs, flash memory is expected to lead the semiconductor industry towards the next revolution in transistor density.
Inspired from this fact, this work focuses on molecular flash memories and logic switching molecular networks which, among all emerging technology candidates, are considered particularly promising due to
their ability for reduction of size per cell and solution processing (low cost, injection-printing friendly), conceptual compatibility with photonic
addressing due to molecular photosensitivity, multilevel storage, high information density, quick write-read operations, low power consumption,
mechanical flexibility, bottom-up fabrication logic (overcoming the lithographic patterning constrains), conceptual non-binary data representation
and properties’ tunability through chemical tailoring.
Molecular electronic devices are fabricated via a combination of bottom-up layer-by-layer nanofabrication and self-assembly with CMOS platform
lithography in order to provide a low cost large-scale route towards extension of the functional value of Si-based platforms.
Tungsten Polyoxometalates (POM, [PW12O40]3- of the Keggin class
are being self-arranged both on nanocrystal and hyperstructure level in
a rational way resulting in layers of tunable spatial correlation length.
The hyperstructures exhibit tunable valence and conduction bands and,
hence, adjustable electronic properties directly related to the extent of
crystallization of their building blocks.
Dimensional crossover-driven insulator-to-semimetal transitions can
be enforced in these hyperstructures via tuning the extent of crystallization in solution. Being able to transport or confine charge at will, these
hyperstructures constitute ideal candidates for alternative molecule-based
solution-printed circuitry components and transistor channels.
Hybrid CMOS/molecular memory devices based on the parallel plate
architecture are fabricated, characterized and tested. Each memory element contains a planar hyperstructure of molecules (typically several millions) that can store charge having multiple times the charge density
of a typical DRAM capacitor.
Transition-metal-oxide hybrids composed of high surface-to-volume
ratio Ta2O5 matrices and tungsten POMs are investigated as a charge
storage composite in molecular nonvolatile capacitive memory cells. Enhanced internal scattering of carriers results in a memory window of 4.0
V for the write state and a retention time around 10 4 s without blocking
medium.
Differential distance of molecular trapping centres from the cells gate
and electronic coupling to the space charge region of the underlying Si
substrate are being identified as critical parameters for enhanced electron
trapping for the first time in such devices.
The incorporation of a molecular-friendly blocking oxide that facilitates long term retention while suppressing cross-talking, is performed
through realization of a multi-functional oxide stack (SiO 2 /hybrid Ta2O5-POM transition metal oxide/Al 2 O 3 ) that takes parallel advantage of photonically-addressed phononic modes to boost information storage and reach
molecular states that were previously non-available. A 37 % information
density increase is attained via phononic pumping, while the memory
window reaches 7.0 V, corresponding to ∼ 4 × 10^14 cm^-2 charging nodes
able to carry 65-195 μCb/cm^2 . Ability of multi-state addressing and write
speed of 10 ns are being documented for the packed cell.
The fabricated high performance non-volatile memories are the first
documented CMOS-compatible long term (10 years criterion satisfaction)
retention molecular capacitive cell of its kind.
Following a different approach, brain-inspired, neural systems performing in networks and data-centric non-Von Neumann processing are
among the latest trends for non-conventional approaches in the semiconductor industry. We focus on hybrid molecular-nanoparticle networks
that exploit the massive parallelism of designless interconnected networks of locally active components, obviating the need for expensive
lithographic steps.
Molecular multi-junction networks comprising of gold nanoparticles
(AuNPs) of diam.∼1.4 nm, electronically linked by means of copper 3-diethylamino-1-propylsulphonamide sulfonic acid substituted phthalocyanine (CuPcSu) molecules are fabricated and studied.
When electrons flow through the non-linked nanoparticle arrays, they
experience on-site Coulomb repulsion and are strongly localized, with localization length (ξ=0.7 nm). Under dynamic excitation the system undergoes Coulomb oscillations, while the introduction of CuPcSu molecules
results in the formation of a network of multiple molecular/Au nanojunctions and conductance increases by 5 orders of magnitude. This switching behaviour functions on reversible red-ox reactions and
pushes carriers in a weak localization state. In this state electrons spread
over several junctions and all temperature scaled current vs voltage curves,
J/T^(1+α) vs eV/kT, collapse in one universal curve, characterizing the network and the extent of its disorder.
On the other hand, the strongly non-linear I-V response and negative
differential resistance of drop-cast nanojunction 3-d arrays makes them
suitable platforms for logic function exhibition. Common miniaturization bottlenecks such as capacitive crosstalk, are embraced as exploitable
physical processes, that can lead to robust computational functionality.
The networks can be configured on-flight with pulses as quick as 10 ns
to modify their resistance between two discrete levels. Both levels can be
addressed real time utilizing patterned nano-electrode pairs and reading
voltage of the order of 500 mV. The networks are able to perform as a two-input “then-if” logic gates.
(EN)
Εθνικό Μετσόβιο Πολυτεχνείο
