Personal tools

Difference between revisions of "Finite element modeling of electrochemical random access memory"

From iis-projects

Jump to: navigation, search
(Created page with "==Short Description== IBM Research –Zurich has an opening for a Master Thesis in the area of finite element modeling of electrochemical devices for analog comput...")
 
Line 4: Line 4:
 
Electrochemical  random  access  memory  (ECRAM)  isa  novel  type  of non-volatile  memory (NVM) [1,2] to  provide  acceleration  for  training  of  deep  neural  networks  (DNNs) [3].  ECRAM is based  on the reversible  electrochemical  intercalation of  an  ion  into  a  host  material,  which causes  a  change  in  electrical  resistance. Compared  to  other  NVMs  such  as  phase-change memory  (PCM)  or  resistive  random  access  memory  (ReRAM),  ECRAM  providesmoresymmetric    and    deterministic    potentiation    and    depression.    However,    a    thorough understanding  of  ECRAM  operation  and  the  influence  of materialspropertiesand geometry on performance metrics requires a detailed multiphysics model
 
Electrochemical  random  access  memory  (ECRAM)  isa  novel  type  of non-volatile  memory (NVM) [1,2] to  provide  acceleration  for  training  of  deep  neural  networks  (DNNs) [3].  ECRAM is based  on the reversible  electrochemical  intercalation of  an  ion  into  a  host  material,  which causes  a  change  in  electrical  resistance. Compared  to  other  NVMs  such  as  phase-change memory  (PCM)  or  resistive  random  access  memory  (ReRAM),  ECRAM  providesmoresymmetric    and    deterministic    potentiation    and    depression.    However,    a    thorough understanding  of  ECRAM  operation  and  the  influence  of materialspropertiesand geometry on performance metrics requires a detailed multiphysics model
  
[1].Fuller, E. J. et al.Li-Ion Synaptic Transistor for Low Power Analog Computing. Adv. Mater.29, 1–8 (2017).
+
[1] Fuller, E. J. et al.Li-Ion Synaptic Transistor for Low Power Analog Computing. Adv. Mater.29, 1–8 (2017).
  
[2].Tang, J. et al.ECRAM as Scalable Synaptic Cell for High-Speed, Low-Power Neuromorphic Computing. in IEEE International Electron Devices Meeting (IEDM)(2018).
+
[2] Tang, J. et al.ECRAM as Scalable Synaptic Cell for High-Speed, Low-Power Neuromorphic Computing. in IEEE International Electron Devices Meeting (IEDM)(2018).
  
[3].Gokmen, T. & Vlasov, Y. Acceleration of deep neural network training with resistive cross-point devices: Design considerations. Front. Neurosci.10, 1–13 (2016).
+
[3] Gokmen, T. & Vlasov, Y. Acceleration of deep neural network training with resistive cross-point devices: Design considerations. Front. Neurosci.10, 1–13 (2016).
  
 
==Objectives & Methodology==
 
==Objectives & Methodology==

Revision as of 18:15, 3 September 2019

Short Description

IBM Research –Zurich has an opening for a Master Thesis in the area of finite element modeling of electrochemical devices for analog computing. The work will be carried out in the Science & Technology Department at IBM Research –Zurich and willinvolve multiphysics modeling and collaboration with experimental research teams in Zurich and the IBM T.J. Watson Research Center in Yorktown Heights, NY, USA.

Electrochemical random access memory (ECRAM) isa novel type of non-volatile memory (NVM) [1,2] to provide acceleration for training of deep neural networks (DNNs) [3]. ECRAM is based on the reversible electrochemical intercalation of an ion into a host material, which causes a change in electrical resistance. Compared to other NVMs such as phase-change memory (PCM) or resistive random access memory (ReRAM), ECRAM providesmoresymmetric and deterministic potentiation and depression. However, a thorough understanding of ECRAM operation and the influence of materialspropertiesand geometry on performance metrics requires a detailed multiphysics model

[1] Fuller, E. J. et al.Li-Ion Synaptic Transistor for Low Power Analog Computing. Adv. Mater.29, 1–8 (2017).

[2] Tang, J. et al.ECRAM as Scalable Synaptic Cell for High-Speed, Low-Power Neuromorphic Computing. in IEEE International Electron Devices Meeting (IEDM)(2018).

[3] Gokmen, T. & Vlasov, Y. Acceleration of deep neural network training with resistive cross-point devices: Design considerations. Front. Neurosci.10, 1–13 (2016).

Objectives & Methodology

An existing multiphysics model created in COMSOL Multiphysics®is availableas starting point. The model has been validated against data reported for thin-film Li-ion batteries. The objective of this project is to compare the model results with in-house experimental data for Li-based ECRAM devices and refineand enrich the model physics to account for experimental observations. Emphasis will be to model the transient concentration profiles during operationand derivekey device metrics. Ultimately, the model shall be used to recommend device configurations in terms of geometry, operating conditions and materialsto enhance performance metrics. The student will work together with our teams in Zurich and Yorktown Heights. Basic knowledge of finite element modeling is an advantage.

Status: Available

Looking for 1 Master student
Interested candidates please contact: Dr. Patrick Ruch
ETH Contact: Mathieu Luisier


Professor

Mathieu Luisier

↑ top