UNLIMITED
POSSIBILITIES
HIGH-PERFORMANCE
SCALABLE SINGLE OR MULTI-CORE
DSP, AI OR AI+DSP SOLUTIONS
USER IN FULL CONTROL OVER
POWER, PERFORMANCE AND AREA
GIVING COST OPTIMIZED IMPLEMENTATIONS
THREE TYPES OF CORES
AI Only
Use as a single core or with multiple cores.
Each individual core can be configured for
number of MACs, quantization and
local TCM memory size
DSP Only
Use as a single core or with multiple cores.
Each individual core can be configured for
number of ALUs, quantization and
local TCM memory size
AI + DSP Combined
Designed for applications where you need to combine the use of AI and DSP with minimal latency and power, for example, ADAS and autonomous driving.
MULTI-CORE ARCHITECTURE
Designed to work as a companion platform
- Fully scalable
- Single core or multi-core
- Algorithm and host processor agnostic
- Fully programmable
Unlimited number of cores
Cores scalable
- Processing power: # of ALUs and/or MACs
- Computation accuracy: IEEE754 floating point
- User selects # of bits in exponent and mantissa
- TCM (local memory) size
- # of DMAs / AXI I/Os
Cores interact through an AXI matrix
Fundamental
New Architecture Type
Does not suffer from external data bandwidth limitations
High Processing Power
- Single core up to 64k MACs (256 TOPS)
- Signals (matrices) handled in hardware (1D / 2D / 3D)
- Signal memory bandwidth follows MPU processing power
- Multi-instructions per cycle / Rich set of instructions
High Flexibility
- High level programming (Matlab-like/Tensorflow-like/C++)
- Fully programmable activation function
- No timing impact from activation function
- Handles algorithms not suitable for parallelism (Cholesky, SVD, ...)
Interested?
We would love to tell you more about our unique solutions and how these will benefit you.
Send us a mail to info@vsora.com
We will take it from there
Send us a mail to info@vsora.com
We will take it from there