Manufacturing Bits: May 6

Ionic memory
Sandia National Laboratories, Stanford University and the University of Massachusetts at Amherst have developed an ionic floating-gate memory array (IFG) for neuromorphic computing.

For some time, the industry has been working on neuromorphic computing. The goal of neuromorphic computing is to replicate the brain in silicon. In a neuromorphic chip, the goal is to mimic the way that information is moving from one group of neurons to another using precisely-timed pulses.

There are several efforts in this arena. For example, Sandia’s IFG technology makes use of a novel redox transistor. The transistor is connected to a conductive-bridge memory, which is a type of resistive RAM (ReRAM).

It also relies on polymers that use ions to store information, and not just electrons, according to Sandia. The programming of the transistor array is done in parallel. “Synaptic weight read-out with currents <10 nanoampere is achieved by diluting the conductive polymer in an insulating channel to decrease the conductance,” according to Sandia in the journal Science Magazine. “The redox transistors endure >1 billion ‘read-write’ operations and support >1 megahertz ‘read-write’ frequencies.”

“With the ability to update all of the data in a task simultaneously in a single operation, our work offers unmistakable performance and power advantages,” said Sandia researcher Elliot Fuller. “This is projected to improve machine learning while using a fraction of the power of a standard processor and 10 times higher speed than the best digital computers.”

Visual cognitive computer
In a recent scientific journal, Vicarious, a startup that is developing artificial intelligence (AI) for robots, has described more details about its efforts to develop a visual cognitive computer (VCC).

VCC is a computational framework that deals with the manipulation challenges in robotics, that is, moving and arranging objects in set places. VCC also brings the industry closer to robotic systems “that have interpretable representation and common sense,” according to Vicarious in a recent issue of Science Robotics.

Vicarious hopes to bring AI to the next level in the field of robotics. “We developed a visual cognitive computer (VCC) architecture that formalizes the cognitive and neuroscience insights about concepts,” according to the startup on its Web site. “People utilize their prior knowledge to learn new concepts—our program induction method works in a similar manner.”

Founded in 2010, Vicarious has received venture capital funding from various companies and executives. The company has also published various papers, including on the subject of sensorimotor contingency theory.

More recently, the company has described more details about its VCC architecture. According to Science Robotics, the VCC consists of the following components–a machine vision system, a model for interactions between objects, an attention controller, an imagination blackboard, and a limb controller.

The technology has demonstrated the ability to arrange different objects on a table top. For example, there are various objects on a table, including a red square, blue square, red round object and a star.

The company has developed induction methods that enables a robot to learn programs and arrange objects. For example, the system will move all objects to the top or will move the red objects to the top. “We use a model to predict which programs are likely given the images for a particular concept, and then update this model as more programs are learned. As the system learns more concepts, it guides the search for even more complex concepts. Our method learned 535 out of the 546 concepts with a search budget of 3 million programs,” according to Vicarious.

Synthetic data
AI.Reverie, a provider of synthetic data to train machine learning algorithms, recently announced a partnership and an investment from In-Q-Tel (IQT). IQT is the venture capital arm of the U.S. Central Intelligence Agency (CIA) and other defense agencies.

There is a demand for AI technology that can deal with the explosion in data in systems. Deep and machine learning are gaining momentum in this arena. But at times, this technology can be proprietary, expensive, and time consuming to manually prepare, according to AI.Reverie.

In response, AI.Reverie recently launched a simulation platform to improve machine learning algorithms. The technology makes use of synthetic data, which is data created in a virtual world rather than collected from the real world.

This in turn reduces cost. The technology creates what the company calls “photorealistic virtual worlds,” which closely mimics a true location where clients’ services are used. “Photorealism ensures that the synthetic data is effective in training AI to operate in the physical world,” according to the company. “Because simulations are easier to control, these virtual worlds are the most effective place to test, train and improve computer vision algorithms.”

“AI.Reverie’s platform enables U.S. government agencies to significantly advance their computer vision capabilities,” said Peter Bronez, senior member of the technical staff at IQT. “It has the potential to significantly improve the accuracy of AI tools that detect objects and activity to keep our nation safe.”