IP Library Granted Patent US 8,949,550
Granted Patent B2
US 8,949,550 · App. 13/635,620 · Granted Feb 3, 2015

Memory-centered communication apparatus in a coarse grained reconfigurable array

Inventors: Ki Young Choi (Seoul, KR); Kyung Wook Chang (Daejeon, KR); Jong Kyung Paek (Seoul, KR)
Assignee: SNU R&DB Foundation
G06F15/7871
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 8,949,550
App. No.
13/635,620
Granted
Feb 3, 2015
Kind
B2
Abstract

The present invention relates to a coarse-grained reconfigurable array, comprising: at least one processor; a processing element array including a plurality of processing elements, and a configuration cache where commands being executed by the processing elements are saved; and a plurality of memory units forming a one-to-one mapping with the processor and the processing element array. The coarse-grained reconfigurable array further comprises a central memory performing data communications between the processor and the processing element array by switching the one-to-one mapping such that when the processor transfers data from/to a main memory to/from a frame buffer, a significant bottleneck phenomenon that may occur due to the limited bandwidth and latency of a system bus can be improved.

Claims (19)

1. A memory-centralized communication apparatus in a coarse-grained reconfigurable array (CGRA), the memory-centralized communication apparatus comprising:

at least one processor;

a processing element (PE) array comprising a plurality of PEs, and a configuration cache where commands being executed by the PEs are saved; and

a central memory comprising a plurality of memory units forming a one-to-one mapping with the processor and the PE array, and performing data communications between the processor and the PE array by switching the one-to-one mapping;

wherein the central memory is used as a scratch pad memory of the processor and is used as a frame buffer of the PE array,

wherein the PE array further comprises an address generating unit that generates a full address to the central memory using arguments received from the PEs.

2. The memory-centralized communication apparatus of claim 1 , further comprising a network/bus interface providing an interface between the central memory and an external device.

3. The memory-centralized communication apparatus of claim 2 , wherein the central memory comprises:

a plurality of memory units each having a single port;

an exclusive region comprising a cross-bar switch switching one-to-one mapping between each of the memory units, and the processor, the PE array, and the network/bus interface; and

a common region realized as a register having multiple ports, where one-to-one mapping information between each of the memory units, and the processor, the PE array, and the network/bus interface is saved.

4. The memory-centralized communication apparatus of claim 3 , wherein the processor, the PE array, or the network/bus interface changes the one-to-one mapping information to perform data communications therebetween, and

wherein the cross-bar switch switches the one-to-one mapping between each of the memory units, and the processor, the PE array, and the network/bus interface, according to the changed one-to-one mapping information.

5. The memory-centralized communication apparatus of claim 4 , wherein the memory units use different physical memory but share the same logical address space.

6. The memory-centralized communication apparatus of claim 3 , wherein the common region saves control signals transmitted between the processor, the PE array, and the network/bus interface, or data having a size less than the size preset by a user.

7. The memory-centralized communication apparatus of claim 3 , wherein each of the memory units is divided into a plurality of blocks, and each of the blocks corresponds to a column or a row of the PEs, and has a bandwidth equal to the bandwidth of a data path of the PEs.

8. The memory-centralized communication apparatus of claim 1 , wherein the PE array further comprises an address generating unit where a table of base addresses to data being accessed by the PEs is saved.

9. The memory-centralized communication apparatus of claim 8 , wherein, if a reading or recording request, an index argument, and an offset argument are received from one of the PEs, the address generating unit finds a base address in the table by using the index argument, generates a full address by adding the base address to the offset argument, and transmits the reading or recording request and the full address to the central memory.

10. The memory-centralized communication apparatus of claim 1 , wherein the processor comprises a cache, and wherein, when a memory access request is issued by the processor, the processor determines whether to access the central memory or the cache, by comparing a requested address to an address range of the central memory.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Sep 17, 2012
From: CHOI, KI YOUNG; CHANG, KYUNG WOOK; PAEK, JONG KYUNG
To: SNU R&DB FOUNDATION
Reel/Frame 028972/0083 →
Priority Claims (1)
KR 10-2010-0023414 · Mar 16, 2010 · national
Continuity (1)
Related Publication 20130024621A1 · Jan 24, 2013