Efinix designed the Quantum® architecture to be the leading programmable technology delivering the best power, performance, and area for silicon products. Our goal is to provide the most flexible programmable technology, tailored for compute-intensive applications including compute acceleration, machine learning, and deep learning.
The basic building block of the Quantum® fabric is the eXchangeable Logic and Routing (XLR) cell. An XLR cell can function as a LUT-based logic cell and a routing switch encoded with a scalable, flexible routing structure. Software decides whether the XLR cell is logic or routing.
Having dedicated logic and dedicated routing seems like a good idea on the surface. In reality it means that you are leaving something on the table, area-wise. Most complex logic blocks have only so many ways to enter and leave the block. Once you have used up all of the doorways, you cannot access any remaining logic in the block. Because any XLR cell can be routing, software can choose the best paths to avoid congestion.
XLR cell advantages
Traditional FPGAs are designed with a fixed ratio of logic elements to routing resources. This ratio is critical because too few routing resources means that it is impossible to connect the logic elements and the FPGA is underutilized. Too many routing resources means that silicon area is wasted, driving up chip costs.
Because of the XLR cell, Efinix FPGAs are smaller, use lower power, and are more efficient than traditional FPGAs. And, unlike traditional FPGAs, Efinix® FPGAs are manufactured on a standard CMOS silicon process. With a standard process, these FPGAs are easier for the fab to produce and they can optimize their production at volume.
The second-generation Quantum® compute fabric layers on even more compute and routing capabilities. On top of the streamlined XLR cell, the Quantum compute fabric has 10K embedded RAM blocks and high-efficiency DSP blocks.
The Quantum compute fabric is made up of configurable tiles, the eXchangeable logic and routing (XLR) cell, that optimizes routing efficiency and speed while achieving high utilization ratios. The fabric also has highly configurable, embedded memory blocks along with dedicated, high-speed, DSP blocks. Together, these features deliver optimum performance for a wide array of applications from edge compute to industrial automation and video processing.