When you have a multi-chip design, eventually you will have to handle the question of interactions between those chips. Following links may come out to be handy.
The Advanced Microcontroller Bus Architecture (AMBA) is used as the on-chip bus in system-on-a-chip (SoC) designs. Since its inception, the scope of AMBA has gone far beyond microcontroller devices, and is now widely used on a range of ASIC and SoC parts including applications processors used in modern portable mobile devices like smartphones.
AMBA is a registered trademark of ARM Limited, and is an open standard, on- chip interconnect specification for the connection and management of functional blocks in a System-on-Chip (SoC). It facilitates right-first-time development of multi-processor designs with large numbers of controllers and peripherals.
AMBA was introduced by ARM Ltd in 1996. The first AMBA buses were Advanced System Bus (ASB) and Advanced Peripheral Bus (APB). In its 2nd version, AMBA 2, ARM added AMBA High-performance Bus (AHB) that is a single clock-edge protocol. In 2003, ARM introduced the 3rd generation, AMBA 3, including AXI to reach even higher performance interconnect and the Advanced Trace Bus (ATB) as part of the CoreSight on-chip debug and trace solution. In 2010 the AMBA 4 specifications were introduced starting with AMBA 4 AXI4, then in 2011 extending system wide coherency with AMBA 4 ACE. In 2013 the AMBA 5 CHI (Coherent Hub Interface) specification was introduced, with a re-designed high-speed transport layer and features designed to reduce congestion.
These protocols are today the de facto standard for 32-bit embedded processors because they are well documented and can be used without royalties.
Here is a Chinese blog with good description of AMBA protocol. It describes the handshaking well with two nice figures.
The Wishbone Bus is an open source hardware computer bus intended to let the parts of an integrated circuit communicate with each other. The aim is to allow the connection of differing cores to each other inside of a chip. The Wishbone Bus is used by many designs in the OpenCores project.
A large number of open-source designs for CPUs and auxiliary computer peripherals have now been released with Wishbone interfaces. Many can be found at OpenCores, a foundation that attempts to make open-source hardware designs available.
Wishbone is intended as a “logic bus”. It does not specify electrical information or the bus topology. Instead, the specification is written in terms of “signals”, clock cycles, and high and low levels.
This ambiguity is intentional. Wishbone is made to let designers combine several designs written in Verilog, VHDL or some other logic-description language for electronic design automation. Wishbone provides a standard way for designers to combine these hardware logic designs (called “cores”). Wishbone is defined to have 8, 16, 32, and 64-bit buses. All signals are synchronous to a single clock but some slave responses must be generated combinatorially for maximum performance. Wishbone permits addition of a “tag bus” to describe the data. But reset, simple addressed reads and writes, movement of blocks of data, and indivisible bus cycles all work without tags.
The Wishbone page of open-cores is here.
2003: AMBA vs Wishbone - “A word of advise. stay away from AMBA if you care about performance. AMBA sucks when it comes to performance. We used it in one of our design and regret ever using it. If you have more than 2 devices sitting on your bus and if your devices need to use the bus frequently, then try some other bus interconnect. AMBA is good for low speed and low performance IO based devices. It does not hold candle to any other bus when it comes to throughput.”
[We do not have any experience to tell whether that is true.]
On crossbars switch
From Wishbone wiki -
Wishbone adapts well to common topologies such as point-to-point, many-to- many (i.e. the classic bus system), hierarchical, or even switched fabrics such as crossbar switches. In the more exotic topologies, Wishbone requires a bus controller or arbiter, but devices still maintain the same interface.
High-performance routers have the task of transmitting traffic in between the nodes of the Internet, the network of networks that carries the vast amount of information among billions of users. The switch fabric is the key building block of every router, and various switch fabric architectures are used in the market products. The crossbar-based switch fabric architectures (both buffered and unbuffered) offer very high performances and are widely used for high-performance routers. However their cost grows quadratically with the input/output port count, since they require internal crosspoints (and buffers) for every input/output port pair.
Recently, a functional-level design of two novel Network-on-Chip based switch fabric architectures was proposed, Unidirectional NoC (UDN) and Multidirectional NoC (MDN), as a replacement of the buffered crossbar switch fabric architecture. In this thesis, we propose the hardware design and implementation of the aforementioned architectures for the FPGA platform. We further improve the routing and scheduling algorithms of these architectures for feasible hardware design. The synthesis and simulations are carried out over a wide range of switch sizes and traffic scenarios. The simulation results are also validated on the FPGA platform, by generating pseudo-random destination addresses for the packets on LFSR based test modules. The results show that UDN outperforms MDN in terms of throughput, whereas MDN offers greater performance-cost ratio. Both architectures offer scalability, flexibility and high performance, confirming the ideas in the original proposal.
A large increase of the number of devices integrated in a single chip in conjunction with the significant demands of modern applications for performance has led the designers to a system development methodology based on integrating multiple pre-verified intellectual property cores. Yet, design productivity requirements push designers to focus on key micro-architectural solutions to manage more efficiently the scaling of multi-core SoCs as well as to increase the degree of design automation, particularly as rapid prototyping using reconfigurable computing is becoming mainstream. In this paper we present a novel interconnect architecture based on optimized components to efficiently manage SoCs that follow either a multi-core based approach or are built to support SIMD-style applications that can exploit the processing power of a pool of hardware resources; first we analyze the design of a crossbar featuring shared-memory combined input-crosspoint buffering as a solution for efficient implementation of on-chip interconnection; second we describe the design of a load-balancer featuring configurable proportional allocation of on-chip resources and in-order delivery as a solution for efficient scheduling and execution of processing tasks. The main focus of the paper is to describe and evaluate the mechanisms designed to distribute and manage data transfers so as to implement an efficient interconnection of the integrated cores and control access to available (either on-chip or off-chip) resources for the implementation of a number of embedded systems and applications. Each of these challenges is handled by the proposed architecture in an efficient way in terms of performance, cost in silicon and flexibility.
Is that too much for bioinformaticians? Should we switch to our regular channels of #ENCODE or genome assembly?