NeuroMatrix® NM6403 DSP with Vector/Matrix engine

نویسندگان

  • Dmitri Fontine
  • Vladimir Tchernikov
  • Pavel Vixne
  • Pavel Chevtchenko
چکیده

The paper describes the architecture of the NeuroMatrix® NM6403 DSP designed for image processing, signal processing and neural networks emulation [1,2]. The paper includes a brief description of the processor structure and its instruction set. The NM6403 is the first DSP based on NeuroMatrix® Core (NMC) comprises an original 32-bit VLIW RISC processor and a 64-bit SIMD Vector co-processor (VCP). In contrast to other modern general purpose DSPs and microprocessors with SIMD units such as: Texas Instruments c64xx, Intel Pentium MMX, Motorola AltiVec PowerPC G4 and Analog Devices TigerSHARC, the new DSP performs variable bitlength vector/matrix arithmetic, logic and saturation operations. The main NMC operation is matrix by vector multiplication. The NM6403 supports shared memory mode for two 64-bit external data buses. Two byte-width communication ports simplify the multiprocessor systems design. The NM6403 has been designed by RC "Module" (www.module.ru) and produced by Samsung 0.5μm CMOS technology. The peak performance up to 14.400 MMACs (million multiplication and accumulations per second) has been achieved at a 50MHz clock rate, 3.3V operating voltage and PBGA256 package.

برای دانلود رایگان متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Hardware Neural Network with MC431 DSP Board

The paper is a description of the way in which one neural network was built on a Single-DSP PCI Board. The board was designed for software evaluation and system prototyping on NeuroMatrix® NM6403 DSP. In the first part of the paper, we will point out the importance of the hardware neural network in different lines of business and we will explain which these lines are. Further on, we will presen...

متن کامل

Time Parameters of Some Parallel Image Processing Algorithmsof Hardware Complex – TMS320C40+NM6403

For image processing in real-time systems it is necessary to use special hardware. But for these systems to show high performance it is necessary to develop parallel algorithms and software. This software ought to use internal features of hardware thoroughly. The complex described in this paper consists of following parts: q Multi-processor complex of 8 DSP TMS320C40 q Russian DSP NM6403. This ...

متن کامل

Using of microprocessor NM6403 for neural net emulation

This paper presents an approach of large neural net emulation by using new neuroprocessor NM6403, designed by RC Module. Model of neural net, hardware supported by neuroprocessor architecture, is discussed and some key features of NM6403, such us processing of data with variable bit length and two 64-bit external buses, are highlighted. Examples of emulation of neural nets by one neuroprocessor...

متن کامل

Electro-Optical DSP of Tera Operations per Second and Beyond (Extended Abstract)

We present a vector-by-matrix multiplier architecture incorporating the high potential of optical computing within electronics. This architecture stems from advances in optical switching technology, optical communication, and laser on silicon, and overcomes previous bottlenecks implied by the speed of transferring information to the optical vector-by-matrix multiplier. Based on this architectur...

متن کامل

A Low Power DSP Engine for Wireless Communications

This paper describes the architecture and the performance of a new programmable 16-bit Digital Signal Processor (DSP) engine. It is developed specifically for next generation wireless digital systems and speech applications. Besides providing a basic instruction set, similar to current day 16-bit DSP’s, it contains distinctive architectural features and unique instructions, which make the engin...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

عنوان ژورنال:

دوره   شماره 

صفحات  -

تاریخ انتشار 2000