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

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

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

Samsung researchers presented information on an embedded 28nm Magnetic Random Access Memory (MRAM) technology. The device had a write energy of only 25 pJ/bit and active power requirements of 14 mW (read) and 27 mW (write) with a data rate of 54 MB/s. was 30 mm2 with a capacity of 16 MB and very high endurance (> 1 E14 cycle). The abstract states that reducing the MTJ at the 14nm FinFET node resulted in a 33% improvement in area scaling and 2.6x faster read times. Samsung sees MRAM as low-leakage working memory (cache memory) for AI and other data-intensive applications.

The image below shows a cross-sectional TEM view of Samsung’s eMRAM bitcell array embedded in a 14nm logic platform.

IMEC demonstrates a lanthanum-doped hafnium-zirconate (La:HZO) ferroelectric capacitor with a high endurance of 1011 cycles, high final remanent bias (2PR = 30µC/cm2 at 1.8 MV/cm) and reduced wake-up time. The researchers achieved this unique combination of properties by engineering the interfacial oxides of the stack of ferroelectric capacitor materials. This high-performance, scalable, CMOS-compatible ferroelectric capacitor technology will be crucial in enabling embedded and standalone ferroelectric random-access memory (FeRAM) applications. The figure below shows the endurance curves for some of IMEC’s ​​ferroelectric capacitors.

In addition to the use of non-volatile memory for computer system memory, it is also used for neuromorphic computing. Elisa Vianello from CEA Leti gave a tutorial on this topic. At the heart of biological signal processing are two fundamental concepts: event detection and in-memory analog computing. Resistive memory provides a compact solution for storing synaptic weights and RRAMs are non-volatile devices, a feature that matches the event-driven asynchronous nature of the system proposed by the team, resulting in zero power consumption. energy when the system is idle.

Leti used neuromorphic computing to activate a barn owl-inspired object location sensor. A diagram of the sensor is shown below.

Neuromorphic computing co-locates memory and computation, reducing power consumption. This application has time parsimony, that is, information is sent only when new data is available. Most connectivity is local and global connections are rare. For these types of applications, neuromorphic computing can provide robust computing with noisy and unreliable computing elements.

Emerging nonvolatile memories will enable low-power applications for storage as well as new computational methods. Coughlin Associates and Objective Analysis estimate that the market for such memories could grow to approximately $44 billion by 2032. The figure below shows the projected growth in shipped MRAM storage capacity, representing these emerging non-volatile memories, by compared to DRAM and NAND Flash growth projections.

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

#Samsung #MRAM #Imec #FeRAM #CEA #Leti #RRAM #Neuromorphic #Computing #IEDM

Leave a Comment

Your email address will not be published. Required fields are marked *