Generative Adversarial

Networks
(Innovation trigger - 5 to 10
years expected)

• GANs are an approach to generative modeling using deep learning methods, such as convolutional neural networks. They are a way of training a
generative model by framing the problem as a supervised learning problem with two sub-models: the generator model trains to generate new examples,
and the discriminator model that tries to classify examples as either real (from the domain) or fake (generated). The two models are trained together in a
zero-sum game, adversarial, until the discriminator model is fooled about half the time, meaning the generator model is generating plausible examples.

• The generator model takes a fixed-length random vector as input and generates a sample in the domain. The vector is drawn from randomly from a
Gaussian distribution, and used to seed the generative process. After training, points in this multidimensional vector space correspond to points in the
problem domain, forming a compressed representation of the data distribution. The generator model applies meaning to points in a chosen latent space,
such that new points drawn from the latent space can be provided to the generator model as input and used to generate new and different output
examples.

• The discriminator model takes an example from the domain as input (real or generated) and predicts a binary class label of real or fake (generated). The
real example comes from the training dataset. The generated examples are output by the generator model. The discriminator is a normal (and well
understood) classification model. After the training process, the discriminator model is discarded as we are interested in the generator.

• Sometimes, the generator can be repurposed as it has learned to effectively extract features from examples in the problem domain. Some or all of the
feature extraction layers can be used in transfer learning applications using the same or similar input data.

• In this way, the two models are competing against each other, they are adversarial in the game theory sense, and are playing a zero-sum game. Zero-sum
means that when the discriminator successfully identifies real and fake samples, it is rewarded or no change is needed to the model parameters, whereas
the generator is penalized with large updates to model parameters. Alternately, when the generator fools the discriminator, it is rewarded, or no change is
needed to the model parameters, but the discriminator is penalized and its model parameters are updated.

• Generative adversarial nets can be extended to a conditional model if both the generator and discriminator are conditioned on some extra information y. y
could be any kind of auxiliary information, such as class labels or data from other modalities. We can perform the conditioning by feeding y into the both the
discriminator and generator as [an] additional input layer.

• GANs are an exciting and rapidly changing field, delivering realistic examples across a range of problem domains, most notably in image-to-image
translation tasks such as translating photos of summer to winter or day to night, and in generating photorealistic photos of objects, scenes, and people.
GANs have seen much success in this use case in domains such as deep reinforcement learning due to their successful ability to model high-dimensional
data, handle missing data, and provide multi-modal outputs or multiple plausible answers.
Carbon-Based Transistors
(Innovation trigger - 5 to 10 years expected)
• In the 21st century, we are in the nanoscopic era—and demand that computing devices are smaller and smaller every day. Accordingly, we are faced
with the challenge of following Moor’s Law, according to which the processing speed of computer chips should be increased twofold every two years.
Accomplishing this continued growth entails an increase of the number of transistors per square inch, and thereby the power density and the chip
performance as well. At the same time, the price of the item should remain unchanged.

• It is obvious that silicon-based transistors are reaching their size limits. The focus of engineers worldwide is to find a suitable alternative solution. It is
beyond doubt that carbon nanotubes are the most promising material that could replace silicon, because of their many unique properties. One such
property is, for instance, their high carrier mobility.

• In 1998 the researchers from Delft University of Technology along with IBM built a transistor by using a semiconducting single-walled carbon atom,
just 1 nanometer in diameter, that could operate at room temperature. Using this technology in commercial purposes has resulted in many technical
challenges. Making the carbon nanotubes places too much emphasis on the minor details which makes them difficult for mass commercial
production. The difficulty in this aspect is also that the process of making carbon nanotubes results in the nanotubes transistors to be even larger than
silicon-based transistors if the optimal performance is required.

• To overcome these difficulties, IBM researchers have recently made special transistor contacts which carry current through the nanotube transistor.
Cobalt-molybdenum is used for making low resistance contacts that can bond directly to the nanotubes ends, thus saving valuable space.

• The use of cobalt ensures that the bond can be established at a lower temperature. Reducing the reaction temperature allows making contacts and
maintaining their structural integrity of under 20 nanometers gaps between the source and the drain electrodes. Researchers have succeeded in
placing several parallel nanotube wires close to each other, thus increasing the total electrical current. Each nanotube is approximately 1.4
nanometers wide and 30 nanometers long. Higher current provides the fastest charging of the gate of the next device in the circuit.

• Physically, there are two main types of CNTFETs that are being currently studied, differing by their current injection methods. CNTFETs can be
fabricated with Ohmic or Schottky contacts. The type of the contact determines the dominant mechanism of current transport and device output
characteristics. CNTFETs are mainly divided into Schottky barrier CNTFETs (SB-CNTFETs) with metallic electrodes which form Schottky contacts and
MOSFET-like CNTFETs with doped CNT electrodes which form Ohmic contacts. In SB-CNTFETs, tunneling of electrons and holes from the potential
barriers at the source and drain junctions constitutes the current.

• The other type of the CNTFETs takes advantage of the n-doped CNT as the contact. Potassium doped source and drain regions have been
demonstrated and the behavior like MOSFETs have been experimentally verified (Javey, et al., 2005). In this type of transistors a potential barrier is
formed at the middle of the channel and modulation of the barrier height by the gate voltage controls the current.

• IBM engineers managed to reduce the transistor’s total footprint to only 40 nanometers (silicon transistors are about 100 nanometers across). The
nanotubes transistors have a high potential for application in wireless communication devices where the relatively high current needs to flow through
the small area. The microchips based on nanotube transistors could be five times faster and use less energy than the silicon chips. According to IBM,
such chips should be ready for commercial use around 2020.
Ruchika 10301012017 CSE 2