Because of silicon microelectronics (also known as chips or microchips), our smartphones and laptops are small but mighty. These tiny brains power almost all modern devices.
However, this convenience comes with a price. Electronic devices will consume 25% of the world’s total energy by 2030. This is even though most of it is generated by burning carbon-rich fossil fuels.
Silicon chips originate from a design known as CMOS, shorthand for complementary metal-oxide-semiconductor. CMOS silicon chips, which Moore’s Law predicted first in 1975, are rapidly approaching limits in performance and miniaturization. Scientists have searched for electronic materials that can surpass the limitations of Moore’s Law and the constraints of silicon CMOS chip technology for decades.
Scientists Maurice Garcia-Sciveres (DOE) and Ramamoorthy Ramesh (Berkeley Lab) are now developing new microchips that can perform better than silicon and use less energy. They will lead two of the ten recent $54 million projects by DOE to improve energy efficiency in microelectronics production and design over the next three years.
Garcia-Sciveres: Our project – the “Co-Design and Integration of Nanosensors on CMOS” – aims to improve performance by integrating tiny light sensors made with nanomaterials into a conventional CMOS (complementary metal-oxide-semiconductor) integrated circuit. A nanomaterial is a material that has been designed on a small scale, approximately one billionth of a meter.
Although CMOS chips are made from silicon, you can see how much power it uses. Silicon chips could consume a significant amount of energy in a decade. The computing required to drive a self-driving vehicle consumes much more energy than it uses to operate. We want to use less energy or improve performance without using more power. But silicon chips can’t do this because silicon must run at a specific voltage. These physical limitations are costly.
Our project would use nanomaterials, such as carbon nanotubes, as light sensors. These devices are so small they can’t be seen by the naked eye. Nanosensors can add functionality to CMOS chips, increasing their performance.
Although sense is an excellent initial application, carbon nanotubes can be integrated into a chip to serve as transistors and switches that process data. Combining many carbon nanotubes in a silicon chip could result in new types of electronic devices that are smaller, faster, and more efficient than existing technologies.
Our project, “Co-Design for Ultra-Low Voltage Beyond CMOS microelectronics,” aims to discover new physical phenomena that can lead to significant energy savings in computing. This is important as the next Moore’s Law will likely be focused on energy and not length since we have already reached the limits of length scaling.
Microelectronics accounted for only 4-5% of total primary energy worldwide in 2015. Primary energy is typically the chemical energy produced by a power plant using natural gas or coal. This energy is converted to electricity at a 35-40% rate.
The systems perspective will see an exponential rise in electronics due to our increasing dependence on artificial intelligence, machine learning, and IoT.
Microelectronics will account for at least 25% of primary electricity by 2030. It is, therefore, essential to make electronics more efficient.
Our project asks: “What fundamental material innovations could significantly reduce the energy consumption for microelectronics?”
Garcia Sciiveres: Our work will show a single-photon imaging device that can measure the spectrum (the wavelength or energy) of each photon or light particle it detects. Hyperspectral imaging is possible, thanks to this. Images, where each pixel can be broken down into multiple colors, provide more information. This technology is used in many scientific fields, including cosmology and biological imaging.
Berkeley Lab manages the Dark Energy Spectroscopic Experiment, an international scientific collaboration that captures the spectra of distant galaxies. It starts from images of galaxies taken previously with other instruments. Cosmologists can use this additional spectral information to help them understand the role of dark energy in expanding our universe. If the original observations of galaxies had been made using hyperspectral imaging, then spectral data would have been available.
Hyperspectral imaging has a growing use in the study of exoplanets. (Planets orbit the Sun around their solar system. Exoplanets are planets that orbit around other stars.
However, the sensors used to make these observations can only be operated at temperatures below 1 degree above absolute 0. The device could work at room temperature or even lower temperatures.
Many commercial instruments and medical devices are available for hyperspectral imaging. There are many uses for hyperspectral imaging in medicine and biosciences. These instruments are more complicated and more costly than regular cameras. They scan objects pixel-by-pixel or use complex combinations of robotic fibers and filters. These instruments are not sensitive to single photons. Our device could be used to create a simple camera that can produce hyperspectral images using single-photon sensitivities.
Our team was created to show the viability of our co-design platform, “Atoms to Architecture,” built on two fundamental physical phenomena.
This is a new behavior in ferroelectric transistor architectures. It provides a way to reduce energy consumption in a silicon microelectronics device. A ferroelectric is a material with an electrical dipole or a pair of positive and negative electric charges. It can be switched with an electric field. The second involves low-voltage electric field manipulation of electronic spin using multiferroics, a new class of materials.
We demonstrated in 2014 a magnetoelectric material that converts the charge into a magnetic spin at 5V. This was followed by a collaboration with Intel researchers to show how it could create a new logic-in-memory device called the MESO device. It uses spins for logic operations.
One of our program’s projects will use our magnetoelectric material to study multiferroic materials at 100 millivolts. This will result in a significant reduction in energy consumption. A millivolt is one-thousandth of a volt.
The second project explores the fundamental physics behind a capacitor device. In this case, a ferroelectric layer is placed on top of a standard silicon transistor to increase its energy efficiency. This is known as the “negative capacitance effect.” Our design could enable a microelectronics device to perform memory and logic functions. This fundamentally differs from current chips that perform logic processing and data storage.