Arithmetic Operations within MemristorBased Analog Memory Mika Laiho

Arithmetic Operations within Memristor-Based Analog Memory Mika Laiho, Eero Lehtonen Microelectronics Laboratory, University of Turku © DSI CONFIDENTIAL

Key Points Be able to program the memristor to the reference resistance using the cyclincally programming scheme Ø Computing capability, such as addition, both on negative and positive analog conductance, is demonstrated Ø © DSI CONFIDENTIAL

Digital Memory verse Analog Memory Digital memory (1 -bit information depending on ON/OFF state, including SRAM, DRAM, PCRAM, MRAM, Flash, etc. ) Ø Analog memory (many intermediate states between ON/OFF states) Ø Memoristor can be used for both digital and analog memory Ø © DSI CONFIDENTIAL

Memristor Ø Without programming threshold Linearly programmed with charge flowing through the device – AC readout for memory application – Pose high requirements on R/W cycles – Ø With programming threshold Nonlinear programmed with charge flowing through the device – DC readout possible – © DSI CONFIDENTIAL

Memristor With Programming Threshold Bipolar reversible and nonvolatile switching of nanoscale Ti. O 2 -x devices J. J. Yang et al. , Memristor switching mechanism for metal/oxide/metal nanodevices, Nature Nanotechnology, 2008, 3, 429 -433 © DSI CONFIDENTIAL

Modeling Memristor The current through the memristor The time derivative of the state variable W The window function • ᵅ and ᵝ are fitting constants that are used to characterize the ON state • ᵡ and ᵞ are the fitting constants used to characterize the net electronic barrier when the memristor is switched OFF • a, b, p and q are constants depending on the physical properties of the memristor • w is the state variable of the memristor © DSI CONFIDENTIAL

Simulated Memristor Characteristics Device simulation based on SPICE model © DSI CONFIDENTIAL

Memristor Analog Memory/Computing Circuit VSSR © DSI CONFIDENTIAL

Memory/Computing Circuit Simulation V 1 CT Vin Imem W Ir Circuit simulation based on SPICE model © DSI CONFIDENTIAL

Processing as Summation Programming phase Monitoring phase gm 1 = gm 4 + gm

Processing as Invertor © DSI CONFIDENTIAL

Processing as Universal Addition © DSI CONFIDENTIAL

Conclusions Memristors could be used as analog memories and for computing Ø A two-memristor configuration was proposed to be used as a memory element so that addition operations of both positive and negative numbers could be performed Ø Further study on performing multiplication and division is expected Ø © DSI CONFIDENTIAL

Limitations Control circuits for programming the memristors is too complicated increasing design complexity Ø Many sequences are required leading to slow programming Ø The programming time is unpredictable Ø © DSI CONFIDENTIAL