SHARC
A 32-bit floating-point DSP family from Analog Devices with VLIW instructions
SHARC is a family of floating-point and fixed-point digital signal processors designed by Analog Devices, with the original design dating to about January 1994. It combines a Harvard, word-addressed, VLIW core with serial links to neighbouring chips, and has been used from audio processing to large radar systems.
- Original design dates to about January 1994; architecture introduced in 1999
- Word-addressed: every address points to a full 32-bit word, not a byte
- 48-bit instructions, 32-bit integers and floats, 40-bit extended floats
- Serial links let many SHARC chips work together as a cheap SMP alternative
- Used in audio gear, guided artillery shells and 1000-CPU radar computers
What it is
SHARC, short for Super Harvard Architecture Single-Chip Computer, is a high-performance digital signal processor from Analog Devices that handles both floating-point and fixed-point arithmetic. Introduced in 1999, with the original design dating to about January 1994, it targets signal processing work. Applications span audio processing, single-CPU guided artillery shells, and 1000-CPU over-the-horizon radar processing computers. Chips carry several serial links to nearby SHARC devices, giving a cheap alternative to symmetric multiprocessing. The core is a 32-bit Harvard, word-addressed, VLIW design with 48-bit instruction encoding.
How it came about
The original SHARC design dates to about January 1994, and the architecture was introduced in 1999. Analog Devices built it for signal processing tasks, and gave the chips serial links so several of them could work together cheaply instead of using symmetric multiprocessing. Compiler support came from Analog Devices through CrossCore and Visual DSP++, plus g21k, a now-defunct GCC fork.
Inside the chip
SHARC is a Harvard, word-addressed VLIW processor. Every address points at a whole 32-bit word, so it has no concept of 8-bit or 16-bit values and is neither little-endian nor big-endian. Instructions are 48 bits wide, integers and normal floating-point are 32 bits, and extended floating-point is 40 bits. Code and data normally sit in on-chip memory that the user divides by word size. A special 48-bit register reaches the full instruction word. Instructions can carry a 32-bit immediate, and without one they often run two or more operations at once. There are two delay slots after a jump, hardware loop control for up to six levels, and two full register sets for quick context switches.
Where it went
SHARC appears across signal processing. Audio processing is one area. Military and defence systems use it too, from guided artillery shells with a single CPU to over-the-horizon radar processing computers built from 1000 CPUs. The serial links between neighbouring SHARC chips let designers assemble such multi-processor systems cheaply.
Versions and family
TigerSHARC is a parallel descendant from Analog Devices. Blackfin, appearing from 2000, is the modern replacement. Competing designs named alongside it include Qualcomm Hexagon, Texas Instruments TMS320, and CEVA, Inc. products.
Why it matters
SHARC gave Analog Devices a floating-point DSP line that scaled from small embedded jobs to very large multi-processor systems, since the serial links let many chips cooperate without symmetric multiprocessing. Its word-addressed, VLIW Harvard design is unusual enough that C characters on it are 32 bits wide.
Questions people ask about SHARC
What does SHARC stand for?
Super Harvard Architecture Single-Chip Computer. It is a digital signal processor line from Analog Devices that does both floating-point and fixed-point arithmetic, and the name reflects its Harvard design, where code and data live in separate memory paths.
Why is SHARC neither big-endian nor little-endian?
Because it is word-addressed. Each address points to a whole 32-bit word rather than an 8-bit byte, so the chip has no notion of byte order. A compiler can still pick a convention when it packs smaller values into one word or handles 64-bit data.
How wide are SHARC instructions and data?
Instructions are 48 bits. Integers and ordinary floating-point values are 32 bits, and extended floating-point is 40 bits. On-chip memory is split by the user into regions of different word sizes, for example one 48-bit area for code and one 32-bit area for data.
How do several SHARC chips work together?
Each processor has a good number of serial links to neighbours. Designers chain them to build larger systems cheaply, as an alternative to symmetric multiprocessing. Over-the-horizon radar processing computers have been built from 1000 SHARC CPUs this way.
What replaced SHARC at Analog Devices?
Blackfin, appearing from 2000, is described as the modern replacement. TigerSHARC is a parallel descendant from the same company. Competing DSP families include Qualcomm Hexagon, Texas Instruments TMS320 and CEVA designs.
Good reading and sources
The facts on this page come from the Wikipedia article "Super Harvard Architecture Single-Chip Computer", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryChip rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.
Like SHARC
All digital signal processorsMore from Analog Devices
All of them
🇺🇸 United StatesBlackfin
Hybrid microprocessor and DSP family by Analog DevicesMicrocontroller · RISC · 2000