Samsung MRAM, Imec FeRAM, CEA Leti RRAM Neuromorphic Computing At The 2022 IEDM

The IEEE International Electron Devices Meeting (IEDM) is always a source of interesting information on the latest developments in solid-state technology and in particular solid state memory and storage. Let’s look at some developments in emerging non-volatile memories at the 2022 IEDM.

Samsung researchers presented information on a 28-nm embedded magnetic rando- access memory (MRAM) technology. The device had write energy of only 25 pJ/bit and active power requirements of 14mW (read) and 27mW (write) with a 54 MB/s data rate. The device package was 30 mm2 with 16Mb capacity and very high endurance (>1 E14 cycles). The abstract said that scaling the MTJ down to the 14nm FinFET node resulted in a 33% improvement in area scaling and 2.6X faster read times. Samsung is looking at MRAM as a low-leakage working memory (cache memory) for AI and other applications requiring lots of data.

The image below shows a cross-sectional TEM view of the Samsung eMRAM bit-cell array embedded in a 14nm logic platform.

IMEC demonstrates a lanthanum-doped hafnium-zirconate (La:HZO)-based ferroelectric capacitor with high endurance of 10¹¹ cycles, high final remnant polarization (2P_R = 30µC/cm² at 1.8MV/cm) and reduced wake-up time. Researchers obtained this unique combination of properties by interfacial oxide engineering of the ferroelectric capacitor material stack. This high-performance, scalable, CMOS-compatible ferroelectric capacitor technology will be crucial for enabling ferroelectric random-access memory (FeRAM)-based embedded and standalone memory applications. The figure below shows endurance curves for some of the IMEC ferroelectric capacitors.

In addition to the use of non-volatile memory for computer system memory it is also being used for neuromorphic computing. CEA Leti’s Elisa Vianello gave a tutorial on this topic. At the heart of biological signal processing are two fundamental concepts: event-driven sensing and analog in-memory computing. Resistive memory provides a compact solution to store the synaptic weights and RRAMs are non-volatile devices, a feature that matches the asynchronous event-driven nature of the team’s proposed system, resulting in no power consumption when the system is idle.

Leti used neuromorphic computing to enable a barn owl inspired object localization sensor—a schematic of the sensor is shown below. 

Neuromorphic computing co-localizes memory and computation, which reduces power consumption. This application has temporal sparsity, that is information is sent only when new data is available. Most of the connectivity is local and global connections are sparse. For these types of applications neuromorphic computing can provide robust computing with noisy and unreliable computing elements.

Emerging non-volatile memories will enable low energy applications for storage and also new computing methods. Coughlin Associates and Objective Analysis estimate that the market for these memories could climb to about $44B by 2032. The figure below shows projected growth of MRAM shipped storage capacity, representing these emerging non-volatile memories, compared to projections for the growth of DRAM and NAND flash.

The 2022 IEDM revealed some of the latest developments in solid state storage and memory, including dense low power embedded MRAM from Samsung, high endurance ferroelectric capacitor for FeRAM from imec and object localization using neuromorphic computing with RRAM. We estimate the emerging non-volatile memory market could be $44B by 2032.