Memory subsystem

• US 8,566,515 B2
• Filed: 02/26/2009
• Issued: 10/22/2013
• Est. Priority Date: 01/12/2009
• Status: Expired due to Fees

• Current Assignee: Bring Technology Inc.
• Original Assignee: Maxim Integrated Products, Inc. (Analog Devices, Inc.)
• Inventors: Rubinstein, Jorge, Rooyakkers, Albert
• Primary Examiner(s): Bragdon, Reginald
• Assistant Examiner(s): Ahmed, Hamdy S
• Application Number:
  US12/380,410
• Publication Number:
  US 20100177585A1
• Time in Patent Office: 1,699 Days
• Field of Search: 711/167
• US Class Current: 711/104
• CPC Class Codes: G06F 12/0875 with dedicated cache, e.g. instruction or stack
• CPC Class Codes: G06F 13/1663 Access to shared memory
• CPC Class Codes: G11C 19/00 Digital stores in which the information is moved stepwise, e.g. shift registers
• CPC Class Codes: H04N 19/11 among a plurality of spatial predictive coding modes
• CPC Class Codes: H04N 19/186 the unit being a colour or a chrominance component
• CPC Class Codes: H04N 19/423 characterised by memory arrangements H04N19/433 takes precedence
• CPC Class Codes: H04N 19/43 Hardware specially adapted for motion estimation or compensation
• CPC Class Codes: H04N 19/436 using parallelised computational arrangements
• CPC Class Codes: H04N 19/61 in combination with predictive coding
• CPC Class Codes: H04N 19/82 involving filtering within a prediction loop
• CPC Class Codes: H04N 19/91 Entropy coding, e.g. variable length coding [VLC] or arithmetic coding

Abstract

Embodiments of the present invention are directed to memory subsystems implemented within, or connected to and accessed by, parallel, pipelined, integrated-circuit implementations of computational engines designed to solve complex computational problems. Additional embodiments of the present invention are directed to memory subsystems implemented within, or connected to and accessed by, a variety of different types of electronic devices. One embodiment of the present invention comprises a memory controller implemented in a first integrated circuit or other electronic system and one or more separate memory devices. Alternative embodiments of the present invention incorporate the memory controller within one or more memory devices that are connected to, and accessed by, an integrated-circuit-implemented computational engine or another electronic device. In alternative embodiments of the present invention, the memory controller and memory are together integrated within a computational engine or another electronic device. Alternative embodiments of the present invention include a multi-access memory that interfaces to a simpler memory controller for connection to, or integration within, a computational engine or other electronic device.

Claims

1. A memory subsystem comprising:

a memory; and

a memory controller that

provides one or more data-stream interfaces coupled to receive a data-stream input that includes a plurality of data organized according to a predetermined data structure,

provides a random-access interface to individual memory cells and to two-dimensional memory regions within the memory, each individual memory cell and each two-dimensional memory region being associated with a (x, y) coordinate pair that describes a corresponding location within the memory,

arbitrates among the data-stream interfaces and random-access interface to serialize concurrently requested memory accesses received through the data-stream interfaces and random-access interface,

carries out memory accesses requested through the data-stream interfaces by writing the data-stream input to the memory, and

carries out single-memory-cell and two-dimensional-memory-cell-region memory accesses requested through the random-access interface by reading values from, or writing values to, the individual memory cells and the two-dimensional memory-cell regions in the memory.

2. The memory subsystem of claim 1 implemented within a single-integrated-circuit computational engine.

3. The memory subsystem of claim 1 implemented within a computational engine chip set.

4. The memory subsystem of claim 1 implemented within a memory integrated circuit.

5. The memory subsystem of claim 1 wherein the memory subsystem is controlled by a fastclk clock signal with a frequency that is at least n times greater than the frequency of the fastest clock signal received by the memory controller through any of the one or more data-stream interfaces, wherein n is greater than 2 and wherein the clock signals received from data-stream interfaces facilitate synchronization of memory-controller reception of data units of the data streams received through the data-stream interfaces by the memory controller.

6. The memory subsystem of claim 5 wherein the frequency of the fastclk is sufficiently high to allow the memory subsystem to satisfy memory-access requests received through the random-access interface while storing the data-streams input through the one or more data-stream interfaces.

7. The memory subsystem of claim 5 wherein the memory subsystem includes an arbiter that arbitrates among the data-stream interfaces and the random-access interface to serialize concurrently requested memory accesses received through the data-stream interfaces and the random-access interface, the arbiter comprising a data-source arbiter block corresponding to each data-stream interface and a memory-sequencer block corresponding to the random-access interface.

8. The memory subsystem of claim 7 further including one or more data-source blocks that each implements a data-stream interface and that is each associated with a data-source arbiter block for the data-stream interface and a memory-sequencer block that implements the random-access interface that provides access to individual memory cells and to two-dimensional memory regions and that is associated with the memory-sequencer arbiter block.

9. The memory subsystem of claim 8 wherein each data-source block includes a FIFO queue for temporarily storing data units of the data stream input through the data-stream interface implemented by the data-source block.

10. The memory subsystem of claim 9 wherein redundant data units included in a data stream received through the data-stream interface implemented by the data-source block are dropped by the data-source block, rather than inserted into the FIFO queue.

11. The memory subsystem of claim 7 wherein the one or more data-source arbiter blocks and the memory-sequencer arbiter block are linked together in a sequence, the position of the data-source arbiter blocks and the memory-sequencer arbiter block within the sequence defining a priority for each of the data-source arbiter blocks and the memory-sequencer arbiter block, with the data-source arbiter block in a first position having highest priority, the remaining priorities decreasing with increasing position of the associated data-source arbiter blocks in the sequence, and the or memory-sequencer arbiter block having lowest priority.

12. The memory subsystem of claim 11 wherein the memory-sequencer arbiter block acquires control of memory until a potential for data loss from a data stream input through a data-stream interface forces the arbiter to revoke memory control from the memory-sequencer arbiter block and grant memory control to the data-stream interface.

13. The memory subsystem of claim 11 wherein a single data-source arbiter block or the memory-sequencer arbiter block is granted control of memory at any time, wherein higher-priority data-source arbiter blocks associated with data-stream interfaces through which memory access is currently requested prevent granting of control of memory to lower-priority data-source arbiter blocks, and wherein the highest-priority, memory-access-request-receiving data-source arbiter block or memory-sequencer arbiter block is next to be granted control of memory.

14. The memory subsystem of claim 11 wherein each data-source arbiter block acquires control of memory for a single write operation, but then relinquishes control of memory to allow other data-source arbiter blocks or the memory-sequencer arbiter block to acquire control of memory.

15. The memory subsystem of claim 1 wherein the memory comprises one or more memory integrated circuits.

16. The memory subsystem of claim 1 wherein the data-stream interfaces receive data streams from video cameras and the random-access interface receives random-access memory requests from a video codec.

17. A multi-access memory comprising:

a rectilinear grid of memory cells, each memory cell storing a bit of information;

a row demultiplexer that directs bits from a data stream into memory cells, by using data-transfer shift operations, the bits organized in the data stream according to a predetermined data structure and being transferred to a row of memory cells selected by the row demultiplexer within the rectilinear grid of memory cells; and

a row-demultiplexer-and-column-demultiplexer pair that directs a bit received in a random-memory-access WRITE request into a memory cell selected by the row-demultiplexer-and-column-demultiplexer pair and that retrieves a bit requested by a random-memory-access READ request from a memory cell selected by the row-demultiplexer-and-column-demultiplexer pair, each request of the WRITE and READ requests being associated with a (x, y) coordinate pair that describes the address of the memory cell within the rectilinear grid of memory cells.

18. The multi-access memory of claim 17 wherein the multi-access memory includes one or more row demultiplexers that each directs bits from a data stream into memory cells by shift operations along a row of memory cells selected by the row demultiplexer, each row demultiplexer associated with a different region of the rectilinear grid of memory cells.

19. The multi-access memory of claim 18 wherein the multi-access memory is simultaneously accessed for writing at least one row, by shift operations, from a data stream and by a random-memory-access WRITE request and a random-memory-access READ request.

20. The multi-access memory of claim 17 wherein a memory cell includes an inverting data-storage unit that stores a complement of a data bit, received on a data-storage-unit input, statically, without update and that continuously output the stored complement of the data bit to a memory-cell shift output.

21. The multi-access memory of claim 20 wherein the data-storage unit is a flip flop comprising two inverters.

22. The multi-access memory of claim 21 wherein the memory cell further comprises:

a first shift-operation transistor, a source input of which is connected to a memory-cell data-stream input and a gate of which is connected to a memory-cell Shift1 input;

a second shift-operation transistor, a drain output of which is connected to the data-storage-unit input and a gate of which is connected to a memory-cell Shift2 input;

an inverter connected, an input of which is connected to a drain output of the first shift-operation transistor and an output of which is connected to a source input of the second shift-operation transistor;

a memory-cell-random-access output transistor, a gate of which is connected to a memory-cell memory-cell-random-access-read input, a source input of which is connected to the memory-cell shift output, and a drain output of which is connected to a memory-cell random-access read output; and

a memory-cell-random-access input transistor, a gate of which is connected to a memory-cell memory-cell-random-access-write input, a source input of which is connected to a memory-cell random-access input, and a drain output of which is connected to the data-storage-unit input.

23. The multi-access memory of claim 22 wherein a bit is written into the memory cell from the memory-cell random-access input by raising the memory-cell random-access-write input.

24. The multi-access memory of claim 22 wherein a bit read from the data-storage unit on the memory-cell random-access read output by raising the memory-cell memory-cell-random-access-read input.

25. The multi-access memory of claim 22 wherein, in the rectilinear grid of memory cells, rows of memory cells function as shift registers, with the memory-cell shift output of all but the last memory cell in each row connected to the memory-cell data-stream input of a next memory cell in the row.

26. The multi-access memory of claim 22 wherein a bit is shifted into the memory cell from the memory-cell data-stream input by raising the Shift1 memory-cell input, lowering the Shift 1 memory-cell input, and raising the Shift2 memory-cell input.

27. The multi-access memory of claim 17 coupled to a memory controller to compose a memory subsystem.