Philips 75003407 DATASHEETS

The TriMedia™ TM-1000 is a general-purpose microprocessor for
real-time processing of audio, video, graphics, and communications
datastreams. In a single chip, TM-1000 combines an ultra-high
performance, low cost CPU with a full complement of I/O and
coprocessing peripheral units.

In consumer electronics appliances and personal computing prod­ucts, the TM-1000 media processor performs multimedia functions
with the advantages of special-purpose, embedded DSP solutions —
low cost and single-chip packaging — and the programmability of
general-purpose CPUs. It improves time-to-market through high­level C/C++ language programmability and delivers throughput of
up to four billion operations per second.

MUL TIMEDIA APPLICATIONS

TM-1000 is an ideal building block for any multimedia application
that requires processing of video, audio, graphics, and communica­tions datastreams. It is well suited for applications ranging from
single-purpose systems such as videophones, to reprogrammable,
multipurpose devices such as set-top boxes or web browsers.

TM-1000 easily implements popular multimedia standards such as
MPEG-1 and MPEG-2, but its orientation around a powerful, pro­grammable general-purpose CPU makes it capable of a variety of
multimedia algorithms, whether open or proprietary.

HARNESSING THE POWER OF VLIW

TM-1000 delivers top performance through its elegant implementa­tion of a fine-grain parallel architecture known as very-long instruc­tion word, or VLIW. Unique to the TriMedia processor’s VLIW
implementation, parallelism is optimized at compile time by the
TriMedia compilation system. No specialized scheduling hardware is
required to parallelize code during execution. Hardware saved by
eliminating complex scheduling logic reduces cost and allows the
integration of multimedia-specific features.

With the capacity to pack multiple operations into one VLIW
instruction and 27 functional units in which to process them,
TM-1000 can execute up to five operations in parallel with each
clock cycle. Such parallel processing is an ideal complement to the
inherently parallel nature of multimedia applications.

Another key contributor to TM-1000’s top performance is its use of
conditional execution. During program creation, an instruction
scheduler adds conditional code to each operation to enable guarded
execution — a technique that increases fine-grain parallelism and
significantly decreases code branching and execution time.

FEATURES

+ Processes audio,video,graphics, and communications

datastreams on a single chip

+ Powerful,fine-grain parallel,100 MHz VLIW CPU with

separate instruction and data caches

+ Independent,DMA-driven multimedia I/O units to

format data and multimedia coprocessors to offload
the TriMedia CPU of specific multimedia algorithms

+ High-performance bus and memory system to manage

communication between TriMedia processing units

+ Instruction set includes RISC, multimedia,SIMD-type

DSP,and IEEE-compliant floating point operations

+ Robust software development tools and libraries that

enable multimedia application development entirely in
C/C++ programming languages

+ Configurable for standalone and plug-in card applica-

tions in consumer electronics and PC products

Programmable Media Processor

T riMedia TM-1000

PROGRAMMABLE VLIW CPU

At the heart of the TM-1000 is a powerful DSP-like, 32-bit CPU
core. Its VLIW architecture utilizes a five-issue-slot engine. Parallelism
is achieved by simultaneously targeting up to five of the 27 pipelined
functional units in the TM-1000 processor within one clock cycle.
The most common operations have their results available in one clock
cycle; more complex operations have multi-cycle latencies.

Functional units include integer and floating-point arithmetic units
and data-parallel DSP-like units. They can access 128 fully general­purpose, 32-bit registers during execution. The registers are not sepa­rated into banks; any operation can use any register for any operand.

TM-1000’s instruction set includes common RISC operations, special
DSP operations that perform powerful SIMD functions, custom mul­timedia functions, and a full complement of 32-bit, IEEE-compliant,
floating point operations. Both big and little endian byte ordering are
supported.

The TriMedia CPU provides special support for instruction and data
breakpoints, useful in debugging and program development.

TRIMEDIA INSTRUCTION EXECUTION
TM-1000’s unique VLIW CPU utilizes separate instruction and data caches ,
five issue slots, 27 pipelined functional units, and 128 general-purpose,
32-bit registers to process up to five operations in one clock c ycle.

TM-1000

a single-chip

multimedia

workhorse

First in the family of TriMedia

processors,the TM-1000 is more

than just an integrated micro-

processor with unusual peripherals.

It is a fluid single-chip computer

system controlled by a small

real-time operating system kernel

running on a VLIW CPU.

INSTRUCTION CACHE

SDRAM

INSTRUCTION CACHE

ISSUE SLOT 1 ISSUE SLOT 2 ISSUE SLOT 3 ISSUE SLOT 4 ISSUE SLOT 5

FUNCTIONAL UNITS

DEDICATED INSTRUCTION AND DATA CACHE

TM-1000’s CPU is supported by separate, dedicated on-chip data and
instruction caches. To improve cache behavior and performance, both
caches have a locking mechanism. Cache coherency is maintained by
software.

Data cache is dual-ported to allow two simultaneous accesses. It is
non-blocking, thus handling cache misses and CPU cache accesses can
proceed simultaneously. Early restart techniques reduce read-miss
latency. Background copyback reduces CPU stalls. Partial word (8-bit
and 16-bit) memory operations are supported.

To reduce internal bus bandwidth requirements, instructions in main
memory and cache use a compressed format. Instructions are decom­pressed in the instruction cache decompression unit before being
processed by the CPU.

No external second-level cache is required to deliver media perfor­mance an order of magnitude more than x86 processors.

GLUELESS MEMORY SYSTEM INTERFACE

The TM-1000 memory system balances cost and performance by cou­pling substantial on-chip caches with a glueless interface to synchronous
DRAM (SDRAM). Higher bandwidth SDRAM permits the TM-1000
to use a narrower and simpler interface than would be required to
achieve similar performance with standard DRAM.

TM-1000’s memory interface provides sufficient drive capacity for an
up to 100-MHz, 8-MB memory system (four 2Mx8 SDRAMS).
Larger memories can be implemented by using lower memory system
clock frequencies or external buffers. Programmable speed ratios allow
SDRAM to have a different clock speed than the TM-1000 CPU.
Support for a variety of memory types, speeds, bus widths, and off­chip bank sizes allow a range of TM-1000-based systems to be
configured.

HIGH-SPEED INTERNAL BUS (DATA HIGHWAY)

TM-1000’s internal bus, or data highway, connects all internal func­tion units together and provides access to control registers in each
function unit, to external SDRAM, and to the external PCI bus. It
consists of separate 32-bit data and address buses; bus transactions use
a block transfer protocol. On-chip peripheral units and coprocessors
can be masters or slaves on the bus. Programmable bandwidth alloca­tion enables the data highway to maintain real-time responsiveness in
a variety of applications.

TM-1000 ARCHITECTURE
On a single chip, the TM-1000 incorporates a powerful VLIW CPU
and peripherals to accelerate processing of audio, video, graphics,
and communications data.

Unique to the TriMedia
processor’s VLIW
implementation,
parallelism is
optimized at compile
time by the TriMedia
compilation system.

SDRAM

MAIN MEMORY

INTERFACE

I2C INTERFACE

AUDIO OUT

AUDIO IN

SYNCHRONOUS

SERIAL INTERFACE

TIMERS

VIDEO OUT

VLIW CPU

INSTR.
CACHE

DATA

CACHE

IMAGE

COPROCESSOR

PCI INTERFACE

VIDEO IN

VLD COPROCESSOR

TO
PCI BUS