Nano-scale Electronic and Optoelectronic Devices Based on Two-dimensional Materials (Wenjuan Zhu)
1 Video View·Sep 5, 2023
9/8/2016
MNTL Industry Affiliates Program
Good afternoon. Today I'll discuss the research in my group focused on nanoelectronics and
optical electronics based on 2D materials. Here I will highlight a few results we got
in the past few years on this topic. For graphene, it has very high mobility but has zero band
gap so we use graphene to build RF device both on rigid substrate and flexible substrate.
We were able to demonstrate 350 gigahertz cutoff frequency for graphene on silicon carbide.
That's a world record at that time for graphene transistors. We also successfully demonstrated
graphene RF device on flexible substrate on polyimide. So this is what will be one step
on flexible electronics using 2D materials. For logic transistors, graphene is not suitable
because they have zero band gap. In that case, transition metal dicalcogenide like
molybdenum disulfide will be a better choice which have finite sizable band gap and reasonable
mobility. So we built MOS2 transistors aggressively scaled down the channel length. We found that
the DBOB, drain-induced barrier loading, didn't show an upturn until we shrink down the channel
length up to 32 nanometer. So this is used a very thick gate dielectric. It's a 16 nanometer
hafnium oxide for short channels. If we extrapolate, if we scale down the gate oxide as well to
three nanometer, we can predict that the channel length can scale down below seven nanometers.
This means 2D materials can be used for aggressively scaled logic devices. So that's
one potential applications for 2D materials. The third category of device we made is plasmonic
device. Since graphene has very high mobility, it's similar to metals. We can actually pattern
graphene into nanoribbons, nanodisc, and make them into arrays to behave like a plasmonic device.
So here we use graphene nanoribbon arrays to build photodetectors. So the red line is the
photocurrent with plasmonic feature in it, and the blue line is without plasmonic feature.
That means the light polarization is perpendicular to the ribbon, and the blue line is parallel to
the ribbon. So we can see the photocurrent is much higher if we have localized plasmons.
This localized plasmon comes from the lateral confinement of the nanoribbons. We can see that
the enhancement of the photocurrent can be much higher than 10. This plasmon resonance frequency
can also be tuned by gate bias. That will change the current density in graphene. This means the
plasmonic device built by graphene can be tuned dynamically during operation, even after the
device is made. The graphene plasmonics can also be used for biosensing. So if we use graphene
pattern into nanoribbons and spin PMMA or other molecules, or one atomic layer on top, for example
one monolayer of boron nitride, then we can measure the attenuation of the infrared light.
We found that if we have this nanoribbon feature on top, underneath the material,
we can enhance the signal by about five times. So this can be used for biosensing.
The number fifth type of device that we're investigating is the tunneling transistors.
This uses 2D materials, stacks them together, forms a heterostructure, and forms top gate,
bottom gate, and source stream features. So this transistor will be different from traditional
transistor. Traditional transistor is based on the source and drain. From source and drain,
the diffusive transport. For tunneling device, however, it's based on band-to-band tunneling.
So from source and drain, this band-to-band tunneling will be controlled by the energy
band between source and channel. When these are not perfectly aligned, the tunneling path will
be cut off. So you have a perfect zero current in that case. Then we can make a much steeper
transition from on and off. So this steep, sub-pressure swing will mean that you can use
lower supply voltage. So traditional CMOS, the supply voltage is limited by how much
voltage you need to apply to switch from on to off. If you want the on-off ratio of 10 to the fourth,
then you need at least four decades of current transition. That means you have very finite
voltage you need to apply. The minimum sub-pressure swing for traditional CMOS is 60 millivolts per
decade. That means you need at least 0.24 volt. If you use tunneling FET, however,
this limitation will be removed. Your potential, theoretically, you can go to zero,
this transition. So that means maybe 0.1 volt is enough for your supply voltage.
So we simulate the TFET based on black phosphorus. We found that the switching energy versus
switching delay each have a much better characteristic as compared to traditional
high-performance or low-power CMOS. So this means TFET will have a promising application for
low-power logic. The last type of device we're investigating is ferroelectric memories.
Ferroelectric memory typically either use a ferroelectric capacitor or use a ferroelectric
transistor. That means if you apply electrical field across the ferroelectric materials,
the dipole will switch the direction. Even after you remove the electrical field,
the dipole will remain in that direction. So you have a memory effect.
This ferroelectric memory, random access memory, will have advantage of low power as compared to
SRAM, DRAM, or MRAM. That stands for magnetic random access memory. And PRAM, that's phase
change memory. FRAM also will have similar power consumption as RRAM, that's resistive memory.
However, you have much better endurance as compared to RRAM. So ferroelectric memory
is a very promising non-valid memory. Traditionally, the ferroelectric memory is
built by complex peroxide like PZT or SBT. However, those materials have problem is,
number one, it's not compatible with CMOS. So it's very difficult to introduce this material
into a CMOS line. Number two is, it's hard to scale. It's usually deposited by spin-on process,
so you have a thickness limitation. Below that, you won't get a ferroelectricity.
Recently, a new ferroelectric material was discovered. It's hafnium-based ferroelectric
material. So if you dope aluminum, zirconium, or lanthanum in hafnium oxide, this high-cap
dielectric could be ferroelectric. So this new ferroelectric material has many advantages.
Number one is, it has good scalability. You can scale down below five nanometer, you still get
a very strong ferroelectricity. Recently, I think people even demonstrate 2.5 nanometer
zirconium doped hafnium oxide, so it's decent ferroelectricity. Number two is,
it's fully compatible with CMOS. Right now, for Intel, IBM, TI, TMFC, already using hafnium oxide
as gate dielectric. So it's no problem to introduce this material into a CMOS line.
The only thing you need to change is introduce doping into that material.
For example, aluminum oxide is also commonly acceptable for CMOS production. So the barrier
to introduce this material will be a very low threshold. Additional advantage for this material
is a high cohesive field and long retention time. So recently, we successfully made
ferroelectric transistors with doped hafnium oxide on 2D material, like molybdenum disulfide,
and it shows reasonable polarizations and sizable memory windows. This indicates that we potentially
can combine this ferroelectric material with 2D material, make extremely compact memory cells,
and those memories can be dynamically configurable in between DRAM-like
memories and flash-like memories. For DRAM-like memories, you have a very
long endurance, but very short retention. The data cannot store very long, but you can read
and write many times. For flash-like memories, like flash memory, it has long retention. You can
storage the data for a long time, but it won't have very good endurance. It can only last like 10
to the 5 cycles, then it will die. So you can configure this memory to behave like one of these
types of memories by controlling how much pulse you put in there to write this memory. So you don't
need to decide ahead of time how you're going to design the circuit. You can decide it later
by the customer. So that's an advantage for this type of memories. In addition, it's low power
because it's non-volatile memories. So here I would like to acknowledge my former colleagues in IBM
and my students at UIUC and my collaborators in University of Minnesota, University of Massachusetts,
and many other universities and funding agencies. Thank you very much for your attention.

