Bolsa de PD em Arquitetura de Computadores

Post doctoral fellowship in Computer Architecture

Nº: 629

Área de conhecimento: Ciência da Computação

Field of knowledge: Computer science

Nº do processo FAPESP: 2013/08293-7

FAPESP process: 2013/08293-7

Título do projeto: CEPID Center for Computational and Engineering Sciences

Project title: New memory technologies and their impact in large scale computing

Área de atuação: Arquitetura de Computadores

Working area: Computer Architecture

Pesquisador responsável: Rodolfo Jardim de Azevedo

Principal investigator: Rodolfo Jardim de Azevedo

Unidade/Instituição: Instituto de Computação/Unicamp

Unit/Instituition: Instituto de Computação/UNICAMP

Data limite para inscrições: 15/06/2014

Deadline for submissions: 2014-06-15

Publicado em: 29/05/2014

Publishing date: 2014-05-29

Localização: Instituto de Computação,,

Locale: Instituto de Computação,,

  • Resumo Summary

    A tecnologia dominante para memória RAM em computadores está alcançando seu limite físico de fabricação. Nessa linha, novas tecnologias surgem e têm se mostrado promissoras, como PCM, ST-RAM e MRAM. 

    Uma das características principais dessas novas memórias é serem não voláteis, como os dispositivos de memórias secundárias. 

    O objetivo deste projeto é pesquisar a Memória Universal, dispositivo único que possa unir tanto o armazenamento primário, hoje feito por DRAM, quanto secundário, hoje feito por HDs e Flash. 

    Mais detalhes no resumo em inglês e via e-mail rodolfo@ic.unicamp.br.

    Mais informações sobre o CEPID: http://cepid.fapesp.br/centro/13/.

    The current technologies for computer memory, DRAM, is reaching its end of life in terms of scalability due to the technology process evolution that requires even smaller memory cells. 

    Nowadays, there is no prototype implementation of DRAM cells below 40nm, while it is already common to have integrated circuits using 32nm transistors and prototypes using 22nm. 

    There are several technology candidates to replace DRAM, each of them in a different stage of development, like Phase Change Memory (PCM), Spin-Torque Memory (ST-RAM), and Memristor (MRAM). 

    The main common characteristic of DRAM alternatives is that they are non-volatile memories, capable of holding the data even after power is unplugged. Such characteristic is already available in secondary storage like hard-drives (HD) and solid state drives (SSD). 

    Indeed, PCM is already substituting FLASH in SSD due to its higher durability (millions of writes instead of tens of thousands). Using a non-volatile technology as primary and secondary memories will raise questions on the reason to have two different memories in a system, open path to the Universal Memory, one simple storage that will hold both current and persistent states. 

    But our software is not ready to for such change. This project will research the impact and propose solutions to this problem. The following sub-problems will be studied: 

    • Security: Current programs store data in DRAM considering that all the memory content will be erased when the machine is turned off. By using non-volatile memories, sensitive content like password should be erased from memory before closing the program or turning off the machine;
     
    • Protection: Sensitive information from one process should not be available to other process in the machine, even if the machine was turned off between such uses; 

    • Simplification: By having two memory technologies in current systems, we have two different protection mechanisms, page protection for data in DRAM and file protection for data in disk. What need to be changed to unify those two protection mechanism into one? We also need to simplify the set of layers between user program and storage to remove the Operating System from being a bottleneck; 

    • Operating Systems and Program update: We need to rethink common computer tasks like: rebooting, installing, starting and stopping a program, to handle non-volatile memory.

    We are looking for candidates with strong Computer Architecture background. If you are interested in this fellowship, please send a Letter of Interest and a Curriculum Vitae containing key achievements. Please, send your application by email to rodolfo@ic.unicamp.br.

    To learn more about the Center for Computational Engineering and Sciences: http://cepid.fapesp.br/en/centro/13/.