Design of Multiple Master ASIC Interconnect

INTRODUCTION

Intellectual properties are of much significance in complex SoC designs as they tend to have a huge impact on the complexity, design time and cost. Interfacing these IP cores are of much concern in these days as they play a vital role in achieving lossless and efficient communication. In order to achieve lossless communication two protocols are used. First one is Advanced Extensible Interface protocol (AXI). It has five independent channels (write and read address, write and read data, write response channels). All channels use certain signals for transaction. Since, there are two channels for write and read, the transactions are processed in parallel (ARM, 2011). As a result of this there is an additional area overhead. The second one is Open Core Protocol-International Partnership (OCP-IP) (http://www.ocpip.org/). OCP aims in bringing standardization to simplify the problems of integration. It achieves this by the use of a set of interface signals that are independent of the underlying architecture (Karandikar, 2010). Hence designers can focus on the functional aspects of the design, bus designers on the architecture of bus and system integrators on the whole system architecture (Devadas and Latha, 2011). Hence, system integration becomes more efficient.

AXI PROTOCOL

The Advanced Extensible Interface (AXI) protocol has some unique features. It has separate read and write channels thereby providing quicker operations for low latency slaves and also enables Direct Memory Access. Also it supports separate control, address and data states to make transactions more specific. This protocol is burst-based. All transactions have address and corresponding control command on the address channel that specifies the nature of transaction that has to take place. The data is transferred from master to slave through write channel and slave to master through read channel. There is an additional write response channel from the slave to master to specify it has completed the intended transaction.

OPEN CORE PROTOCOL INTERNATIONAL PARTNERSHIP (OCP-IP)

The OCP establishes a peer to peer interface between two IP blocks that intend to communicate with each other (Wang et al., 2008). Here one core acts as the master and the other as slave. The master is the controlling entity and issues command and data. The slave responds to the commands presented to it, either by accepting data or presenting data to the master (Zhang et al., 2009). In interconnect; we performed read, write and write non-post operations. In addition, the proposed interconnect is configurable such that we can connect many masters and slaves depending upon the requirement. In addition, parameters like address width and data width are configurable.

HARDWARE DESIGN OF THE ASIC INTERCONNECT

The blocks used in the interconnect are AXI master, AXI-OCP converter, OCP master, OCP arbiter, OCP address decoder and OCP slave.

AXI master: AXI protocol specifies AXI master as a component which controls the operations such as, read and writes from and to slave. For the Master, the Slave is a location where it can write or reads data. The master is designed as a FSM with read and writes operations. The FSM has two sub FSMs with read and write happening independently due to separate read and write channels. Each operation is initiated with the asserting of the address signal to the slave with its validity. The master waits to receive the response from the slave to proceed further. After getting the response from the slave the master sends the data to be written to the specific address in the slave or reads data from the specific address in the slave (Weber, 2000). The read data channel conveys the data read and response back to the master. The ASIC interconnect we designed is shown in Fig. 1.

AXI-OCP converter: The converter involves mapping the inputs and outputs of AXI Master to OCP specific protocols. All OCP signals are sampled by the positive edge of the clock. Except for clock, all other signals are unidirectional. The commands from the AXI master are input to the AXI-OCP converter which is converted into OCP protocol and forwarded to slave. The signals coming out from OCP converter are OCP protocol specific which can be directly connected to OCP slave interface (Bhakthavatchalu and Deepthy, 2011). The response signals coming from OCP slave interface is input to the AXI-OCP converter which is returned to AXI master in AXI standards.

Clock bridge: The Bus masters and slaves can operate at different clock frequencies. Clock bridges are added to achieve synchronization.

OCP master: The OCP master is designed as a FSM with simple read and writes operations. OCP Master simultaneously generates MCmd, MAddr and MData unlike AXI Master which waits for response from slave before it issues MData (Chang et al., 2010).

OCP arbiter: Arbiter receives request signals from all the masters. It is implemented using a ring counter to select the masters (Shin et al., 2002). The ring counter design is generic such that it is possible to connect a configurable number of masters.

Fig. 1:	ASIC interconnect design

If the selected master is at idle state it goes to next master else it completes read, write or write non-post operations and then goes to next master. This is done sequentially so each master has to wait for its turn.

OCP address decoder: The output of the Arbiter is given to an address decoder. This is used for selecting one slave from a group of slaves. This address decoder design is also generic such that it is possible to connect a configurable number of slaves. The Master that is chosen by the arbiter can communicate with the slave to which it intends. The MAddr signal contains a Slave ID (SID). The first few MSB bits are reserved for the selection of slave. The ID’s are stored in a memory. The Address decoder checks for the SID with which the master intends to communicate and establishes a connection between that master and the slave.

OCP slaves: Three OCP slaves are designed. First one is a simple memory where master can perform read, write operations. Second slave is a synchronous FIFO. Masters can perform read writes to a fixed address. The FIFO is implemented using a memory and write pointers and read pointers which auto increment on a write and read, respectively. Third slave is a dual port FIFO. It is possible for writes and reads to be issued simultaneously on the two ports. In case writes are issued simultaneously on both the slave ports, the slave performs arbitration between the two masters. Similarly, if in case reads are issued simultaneously on both the slave ports, the slave performs arbitration between the two masters.