Researchers at CEA-Leti and Stanford University have developed the world’s first circuit-integrating multiple-bit non-volatile memory (NVM) chip called Resistive RAM (RRAM). The chip is ten times more energy efficient than standard embedded flash memory, thanks to its low power operation and ultra-fast and energy-efficient transitions between on and off modes. To save energy, smart-sensor nodes turn themselves off. It has silicon computing units, and features new memory resiliency, with capacity at least twice that of existing RRAM.
The chip monolithically integrates two heterogeneous technologies: 18kB of on-chip RRAM on top of commercial 130nm silicon CMOS with a 16-bit general-purpose microcontroller core with 8kB of SRAM. The design of 2.3-bits-per-cell RRAM enables higher memory density (NVM dense integration), yielding better application results – for example, more than double the inference accuracy of a neural network, compared to a 1-bit/cell equivalent memory.
To combat the typical problems of write failures in NVM technologies, the research team created a new technique called ENDURER, which extends the chip’s functional lifetime.
Target applications for RRAM include energy-efficient, smart-sensor nodes to support artificial intelligence on the Internet of Things, or “edge AI”, where non-volatility is becoming essential.
“The Stanford/CEA-Leti team demonstrated a complete chip that stores multiple bits per on-chip RRAM cell. Stored information is correctly processed when compared with previous demonstrations using standalone RRAM or a few cells in a RAM array. This multi-bit storage improves the accuracy of neural network inference, a vital component of AI,” said Thomas Ernst, Leti’s chief scientist for silicon components and technologies.
Professor Subhasish Mitra from the Stanford team added that the chip demonstrates several industry firsts for RRAM technology, including new algorithms that achieve multiple bits-per-cell RRAM at the full memory level, new techniques that exploit RRAM features to make it super-efficient at computing, and new resilience techniques that achieve a useful lifetime for RRAM-based computing systems.
The proof-of-concept chip has been validated for a wide variety of applications, including machine learning, control and security.