THIS PAPER will describe a programmable 16b fixed-point Digital Signal Processor (DSP) with an instruction cycle time of 60ns. A 51mm2 chip was processed using an advanced l.Opm, twin-tub, double level metal, CMOS technology; Figure 1. The processor is an enhanced architecture, incorporating an on-chip instruction cache memory for vector operations. A full custom design methodology was employed to design and verify the circuit design and layout.Unit (DAU) is instruction configurable as a 2-stage multiply/ accumulate or a 1-stage ALU. The multiplyladd signal processing data path consists of a 16 x 16 full precision two's complement parallel multiplier and 36b adder feeding two 36b accumulators. Product bit alignment shifter and saturation protection are provided t o simplify programming. Microprocessor style operations utilize the ALU data path that bypasses the multiplier with 1 6 or 32b data to a 15 function ALU and an 8 function shifter, with 4 shifts to both right and left. A set of conditional accumulator functions perform nonlinear signal operations and assist in the performance of many algorithms.An on-chip memory includes 2048 x 1 6 words of ROM for instructions and fixed coefficients, and 512 x 1 6 words of RAM for variable data. An internal ROM can be replaced with a larger external memory of 64K words for full speed prototyping, or for applications that require either frequent program modification or more memory than provided on-chip. Two addressing units support high-speed, register-indirect memory addressing with post-modification. Four address registers can be used for either read or write addresses t o the RAM without restrictions. Modulo addressing of arbitrary length and memory organization is supported. One address register is dedicated to the ROM for table look-up.double buffering and interfaces with codecs, time division multiplexed data, and to other DSPs in a zero-chip multiprocessor Figure 2 is a block diagram of the DSP. The Data Arithmetic The DSP includes two I/O ports. A Serial I/O port (SIO) provides Technology for Logic and Custom VLSI Applications", I E E E 'Tran, A., et. al., "Device Characteristics of a l.Om CMOS Custom Integrated Circuit Conference; 1986. environment. A Parallel I/O port (PIO) is capable of operatingas a bus master or slave for 8 and 16b interfaces. Communication with DSPs, microprocessors, or other 1/0 peripherals is supported. Both the SI0 and the PI0 include a user-maskable interrupt capability for automatic IjO handling. The processor is packaged in an 84-pin plastic or ceramic chip carrier. Many signal processing algorithms consist of repetitive multiply/accumulate instruction sequences. In principle, three busses (instruction, data and data) and three memories are required to execute these operations in, a single cycle. Because of the repetitive nature of the algorithms, a simpler dual bus architecture was chosen for stores and replays of the repeated instructions from a small 15-word cache memory; Figure 3. This memory structure allows parallel access of ins...
The authors describe the successful post-layout verification of the WE DSP32 digital signal processor and the application of advanced CAD tools. The development work, which went on for 10 months, culminated in the smooth transfer of the DSP32 design into high-volume production. During this period, they were able to diagnose and repair hundreds of design errors. They also caught errors in the test-vector software and fixed them before first silicon. The authors estimate that their verification work saved the DSP32 project a year of silicon debugging. In addition, they were able to enhance the CAD tools themselves and better define their future application.'Now with Picturetel. 56o produce a marketable IC with high performance and advanced design features requires a full-custom circuit design. Unfortunately, this type of design, which can in-T volve complex features such as full-hardware floatingpoint arithmetic, is very difficult to verify because it greatly taxes the capabilities of existing CAD tools. The task is even more difficult if these tools have not been successfully demonstrated beyond a certain size and level of complexity. We faced exactly this challenge in verifylng the design of the WE DSP32 digital signal processor. Although the CAD tools we were to use had been applied successfully to other ICs, none approached the size or complexity of our circuit. The DSP32 is a programmable, general-purpose digital signal processor with 32-bit floating-point arithmetic and a 32-bit data path. The chip implements an extensive 32-bit instruction set, 1.2 contains 152,000 transistors and operates at a 16-MHz clock frequency. It has been fabricated in 1.5-1.1 Leff (effective channel length) NMOS technology and measures 12.70 mm x 6.35 mm. The device has been used effectively in complex applications of digital signal processing such a s speech recognition, high-speed modems, and speech CODECs (coder/decoders) with a very low bit rate, and multichannel signaling systems. We realized early on that the DSP32 would be a n acid test of how well the CAD tools would work. *
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.