Over the last few months I have been talking to experts in programmable logic about several key topics that have shaped, and will continue to shape, the industry. In this column, I will summarize what we have heard and chart some directions for future interviews.
This column is also an opportunity for you to add comments and questions on the topics you would like me to explore in future blogs. Let me know what and/or who you would like to hear from next.
As you may recall, several key topics popped up in my earlier discussions, and it's perhaps easiest to summarize some of these key topics in the form of a short "highlights reel." This will refresh our memories and provide some background for thinking about what topics we should ponder and who we should talk to next.
Steve Trimberger, Xilinx:
The increase in size of the FPGA lookup table (LUT) from four, to five, to now six inputs has facilitated optimization of the FPGA routing and logic mix. Larger logic blocks means that more routing can stay local (many times within the logic block itself). This improves silicon efficiency, reduces power dissipation, and improves performance -- all good things! Expect the trend to larger programmable blocks to continue.
Speculation: The current architecture using LUTs in general may finally run out of gas, causing a new architecture to emerge with a larger "grain" and easier expansion to heterogeneous implementations.
Navanee Sundaramoorthy, Xilinx:
The transition to MCU-oriented control in FPGA applications, coupled with the development of automatic algorithm partitioning, will simplify the high-level design of control-based algorithms. Control algorithms will eventually be defined at a high level, and the design software will be able to select the right target for the algorithm based on the required "control loop time."
Speculation: Even the MCU manufacturers will climb onto the programmable logic bandwagon. We will see many "hybrid" devices that combine one or more MCU blocks with multiple programmable fabric blocks. Development software will automatically partition functions into the appropriate blocks based on the desired performance and power consumption, coupled with resource availability.
Arif Rahman, Altera:
The development of special low-power operating modes to keep the growth in operating power in check will be required to enable continued growth in FPGA logic capacity. Without advanced functions similar to those in typical MCUs (sleep, stop, halt, etc.), FPGAs just can't scale fast enough to keep up with customer demand.
Speculation: The power controls offered with FPGA fabric will evolve to allow easier partitioning of low-power logic into special blocks with features like "sleep," whereby the logic is "turned off" until required, saving significant amounts of operating power. More than just "gating the clock," these modes will also result in dramatic reductions in static power consumption.
Shyam Chandra, Lattice:
The need for centralized power control and the development of standard low-power modes on all devices will drive board-level and rack-level architectures. As power control becomes a pervasive need, every device will require some internal "power control" mechanism, along with multiple low-power operating modes. A central "board level" controller will be needed to simplify the process of determining when each device (or section of a device) is on or off.
Speculation: A very complex state machine will be required to control all these power transitions. This state machine needs to be automatically generated based on the devices on the board, the data flow used to process information on the board, and the required low-power operating modes. (Sounds like a good job for a programmable device!)
Raj Nagarajan, Xilinx:
Making FPGA design transparent to software developers will bring a dramatic growth in the number of FPGA designers. As MCUs become a ubiquitous function on FPGAs, and as things evolve such that software developers can simply treat the FPGA as an MCU with some "extra bits" that are automatically configured based on performance, power, and resource constraints, then the dividing line between FPGA designers and software developers will disappear.
Speculation: When this large influx of designers becomes available, we will see a new industry develop where predesigned functions (in the form of software IP) will bring dramatic efficiency improvements to both design and verification. This will keep the design time for the large-capacity devices we will have in the future from being the gating item in time-to-market.
Sinan Kaptanoglu, Microsemi:
The need for new, parallel, place-and-route algorithms to stop the explosive growth in FPGA "compile" times is now of critical importance. All current place-and-route algorithms use approaches that only work on a single "connection" at a time. As devices grow to 10x their current size, compile times are going to explode. We won't get enough speedup for place-and-route from CPU speedups to overcome this. We need algorithms that can work on multiple "states" of the design in parallel -- hopefully hundreds of states at a time.
Speculation: We may get to the point where traditional PCs and workstations simply cannot perform an efficient "compile" for FPGAs. Maybe we will need the equivalent of a graphics accelerator for FPGA place-and-route. (Maybe it will even use FPGAs! Wouldn't that be cool?)
Now it's your turn to speak up...
What kind of speculations do the above topics get you thinking about? Are there other topics and people you want to hear about in my next round of interviews? Let me know by posting comments here, and I will do my best to put them in my queue. And remember -- engineers don't fight the future, we make it!
本视频基于Xilinx公司的Artix-7FPGA器件以及各种丰富的入门和进阶外设,提供了一些典型的工程实例,帮助读者从FPGA基础知识、逻辑设计概念
本课程为“从零开始大战FPGA”系列课程的基础篇。课程通俗易懂、逻辑性强、示例丰富,课程中尤其强调在设计过程中对“时序”和“逻辑”的把控,以及硬件描述语言与硬件电路相对应的“
课程中首先会给大家讲解在企业中一般数字电路从算法到流片这整个过程中会涉及到哪些流程,都分别使用什么工具,以及其中每个流程都分别做了
@2003-2020 中国电子顶级开发网