Dual-mode operation of application specific integrated circuits

1. A method for operating an integrated circuit chip comprising multiple tiles, the method comprising:

determining a configuration for the tiles of the integrated circuit for execution of a computation;

when the configuration for the tiles satisfies a first criterion, operating the integrated circuit in a first mode, including:

concurrently receiving respective input data for the computation at each of the tiles of the integrated circuit; and

when the configuration for the tiles satisfies a second criterion, operating the integrated circuit in a second mode, including:

at a first time, concurrently receiving respective first input data for the computation at each tile of a first group of tiles of the integrated circuit;

at the first time, storing respective second input data for the computation in each of multiple delay registers, each delay register corresponding to a tile of a second group of tiles of the integrated circuit; and

at a second time, releasing the second input data from the delay registers and receiving the released respective second input data at each tile of the second group of tiles.

2. The method of claim 1 , wherein determining a configuration for the tiles for execution of the computation comprises determining one or more of a number and a percentage of tiles operable for executing the computation.

3. The method of claim 2 , wherein:

the first criterion comprises the number or percentage of tiles operable for executing the computation being less than a threshold number or percentage, and

the second criterion comprises the number or percentage of tiles operable for executing the computation being greater than a threshold number or percentage.

4. The method of claim 1 , wherein determining a configuration for the tiles for execution of the computation comprises determining a number of columns of tiles operable for executing the computation.

5. The method of claim 4 , wherein:

the first criterion comprises the number or percentage of columns of tiles operable for executing the computation being less than a threshold number or percentage, and

the second criterion comprises the number or percentage of columns of tiles operable for executing the computation being greater than a threshold number or percentage.

6. The method of claim 1 , comprising determining whether the configuration for the tiles satisfies the first criterion or the second criterion.

7. The method of claim 1 , wherein operating the integrated circuit in a second mode comprises operating a multiplexer to enable the delay registers.

8. The method of claim 7 , wherein operating the integrated circuit in a first mode comprises operating the multiplexer to disable the delay registers.

9. The method of claim 1 , wherein the first time is a first clock cycle and the second time is the next clock cycle.

10. The method of claim 1 , wherein concurrently receiving respective input data in the first mode comprises receiving a vector of input data at each tile.

11. The method of claim 1 , wherein operating the integrated circuit in a first mode comprises operating each of the tiles of the integrated circuit to process the respective input data.

12. The method of claim 1 , wherein operating the integrated circuit in a second mode comprises:

operating each of the tiles of the first group to process the respective first input data; and

operating each of the tiles of the second group to process the respective second input data released from the delay register.

13. The method of claim 1 , comprising receiving an indication of the first criterion and the second criterion from a compiler.

14. An integrated circuit comprising:

multiple delay registers each connected to a corresponding first bus line;

multiple tiles, each tile of a first group of the tiles connected to a corresponding second bus line, and each tile of a second group of the tiles connected to a corresponding one of the first bus lines; and

a controller configured to:

determine a configuration of the first and second groups of tiles for execution of a computation;

when the configuration for the tiles satisfies a first criterion, control the integrated circuit to operate in a first mode in which respective input data for the computation is concurrently received at each of the multiple tiles via the first and second bus lines; and

when the configuration for the tiles satisfies a second criterion, control the integrated circuit to operate in a second mode, in which:

at a first time, respective first input data for the computation is concurrently received at each tile of the first group of tiles;

at the first time, second input data for the computation is stored in each of the delay registers; and

at a second time, the second input data is released from the delay registers and the respective second input data is received at each tile of the second group of tiles.

15. The integrated circuit of claim 14 , comprising a multiplexer connected to each second bus line.

16. The integrated circuit of claim 14 , wherein the tiles are arranged in a two-dimensional array.

17. The integrated circuit of claim 14 , wherein the integrated circuit comprises an application-specific integrated circuit (ASIC).

18. The integrated circuit of claim 14 , wherein each tile comprises:

multiple computational cells;

a memory coupled to the computational cells; and

controllable bus lines configured to convey data to or from the tile.

19. The integrated circuit of claim 14 , wherein at least one of the tiles is configured to act as the controller.

20. The integrated circuit of claim 14 , comprising a communication interface.