Jesús A. del Alamo

Research

Xtreme Transistors

Microelectronics has transformed human society in the last 60 years like no other technology before. The microelectronics revolution has brought immeasurable improvements in the quality of life for billions of people around the world through its impact on communications, computation, health, transportation, energy, education and entertainment, to name a few areas of human concern. As microelectronics becomes transformed into nanoelectronics, the challenges to maintain this revolutionary progress and the benefits that come with it mount.

Our research group at MIT investigates new transistor designs with the goal of pushing the frontiers of electronics to higher frequencies, higher speed, smaller size, lower power consumption, higher operating temperature or to switch electrical power or amplify electrical signals at higher power levels. This is relevant for applications in computation, communications, signal processing, and energy management. Our research currently emphasizes III-V and III-N compound semiconductors but we are also interested in newer materials such as diamond.

Our research is eminently experimental. We design our transistors and nanoscale devices using our CAD tools and we fabricate them in the clean rooms of MIT.nano, MIT’s nanofabrication facility. We then characterize device operation in our own measurement laboratory. Our research leverages heterostructures and materials acquired from commercial sources or from a wide range of collaborators all over the world.

We also study the reliability of transistors under prolonged electrical, thermal and environmental stress. We build models and carry out simulations in order to understand the underlying degradation physics. For maximum relevance, our reliability research is performed in close connection with industrial partners that supply us with devices.

Lately, we have focused our attention on new electronic devices for artificial intelligence applications. Deep learning has irreversibly changed and drastically improved how we process information. However, the enormous computation time and energy costs to train state-of-the-art neural networks make it evident that the future of artificial intelligence hinges on realizing fast and energy-efficient training processors. The concept of analog computing has been put forward as an alternative. This is based on local and fully-parallel information processing in the analog domain using physical device properties instead of conventional Boolean arithmetic. The performance benefits attained by analog training processors are conditional on a set of highly strict properties that are relatively well understood. A plethora of nonvolatile memory technologies (e.g. phase changing, filamentary, magnetic) have been proposed and their use in analog deep learning applications is under intense research. To date, however, no device technology has been identified that meets all the needed requirements. Our group, in collaboration with colleagues at MIT and beyond, is investigating new nanoscale device technologies based on new material systems that operate under new physical principles with the goal of implementing analog deep learning hardware that meets the accuracy of the digital counterpart at a fraction of the energy cost.