Multi-headed multi-buffer for buffering data for processing

A method of buffering a series of tensors and providing multiple access points to a buffered series of tensors, the method comprising:

• configuring a plurality of configurable units in a serial arrangement, and partitioning an address space of each configurable unit of the plurality of configurable units into corresponding sections, such that the serially arranged plurality of configurable units form a chain of sections;
• writing, at a first end section of the chain of sections, a series of tensors at a clock rate;
• sequentially propagating the series of tensors through individual sections within the chain of sections at the clock rate;
• and
• outputting the series of tensors on a second end section of the chain of sections and on an intermediate section of the chain of sections, the first end section and the second end section being two opposite end sections of the chain of sections, and the intermediate section between the first end section and the second end section of the chain of sections,
• wherein each tensor written at the first end section of the chain of sections is output, without modification, on the second end section of the chain of sections and on the intermediate section of the chain of sections.

An integrated circuit, comprising:

• a plurality of configurable memory units, each configurable memory unit having two or more corresponding sections, the plurality of configurable memory units arranged in a serial arrangement to form a chain of sections of the configurable memory units,
• wherein the chain of sections is configured to receive and write a series of tensors at a first end section of the chain of sections, and sequentially propagate the series of tensors through individual sections within the chain of sections,
• wherein the chain of sections is configured to output the series of tensors at two or more access points in the chain of sections, and
• wherein each tensor, as written at the first end section of the chain of sections, is output at the two or more access points in the chain of sections, without any modification to corresponding tensor data.