May 23, 2019 at 8:29 pmDiscussion: the OpenROAD project intends to solve the task of automating the design of processors
Photo - Pexels - CC BY
According to PWC, the semiconductor technology market is growing - last year it reached the $ 481 billion bar. But its growth rate has recently declined . Among the reasons for the recession are the complexity of device design processes and the lack of automation.
A few years ago, Intel engineers wrote that when creating a high-performance microprocessor, you have to use 100-150 separate software tools ( EDA) The situation may be aggravated in the case of heterogeneous devices, the architecture of which includes several different types of chips - ASIC, FPGA, CPU or GPU. As a result, design errors occur that delay the release of products.
Despite the large number of auxiliary tools, engineers are still forced to do part of the work manually. The authors of the book Advanced Logic Synthesis say that sometimes designers have to write scripts in Skill or Python from two million lines to form libraries with cells .
Scripts are also written for parsing reports generated by EDA systems. When developing a chip using a 22-nm process technology, these reports can take up to 30 terabytes.
They decided to correct the situation and try to standardize the design processes in DARPA. The agency also believes that existing methods for creating chips are outdated. The organization launched the five-year OpenROAD program , the purpose of which is to develop new tools for automating chip design processes.
What kind of program
The program involves several projects that use machine learning and cloud technology to automate the individual stages of chip creation. As part of the initiative , more than ten tools are being developed (Scheme 1). Next, we will talk more about some of them: Flow Runner, RePlAce, TritonCTS, OpenSTA.
Flow Runner is a tool for managing RTL and GDSII libraries. The latter are database files that are the industry standard for exchanging information about integrated circuits and their topologies. The solution is based on Docker container technology. You can run Flow Runner both in the cloud and on-premises. The installation guide is in the official repository on GitHub .
Replce- A cloud-based solution based on machine learning, which is responsible for placing components on a chip and automating tracing. According to some reports , intelligent algorithms increase the efficiency of the tool by 2-10% compared to classical systems. In addition, implementation in the cloud simplifies scaling. The installation and configuration guide also lies in the repository .
TritonCTS is a utility for optimizing clock pulses supplied to the chip. Helps route clocks to all parts of the device with the same latency. The principle of operation is based on H-trees . This approach enhancessignal distribution efficiency by 30%, compared with traditional methods. Developers say that in the future this figure can be increased to 56%. TritonCTS source code and scripts are available on GitHub .
OpenSTA is a static time analysis engine. It gives the developer the opportunity to check the performance of the chip even before its actual assembly. The code sample in OpenSTA looks like this.
@@ -6,7 +6,7 @@ read_liberty -corner ff example1_fast.lib
read_verilog example1.v
link_design top
set_timing_derate -early 0.9
set_timing_derate -early 1.1
set_timing_derate -late 1.1
create_clock -name clk -period 10 {clk1 clk2 clk3}
set_input_delay -clock clk 0 {in1 in2}
# report all corners
The utility supports netlist-descriptions of Verilog code, libraries in Liberty format, SDC files, etc.
Advantages and disadvantages
Experts from IBM and IEEE note that cloud computing and machine learning have long been used in chip manufacturing. In their opinion, the DARPA project can be a successful example of the implementation of this idea and will initiate changes in the industry.
It is also expected that the open nature of OpenROAD will allow for the formation of a powerful community around tools and attract new startups.
Photo - Pexels - CC BY
There are already participants - a laboratory engaged in the development of chips based on the University of Michigan will be the first to test OpenROAD open tools. But it is still unknown whether new solutions will be able to have a noticeable effect on the cost of final products.
In general, it is expected that the tools developed under the leadership of DARPA will have a positive impact on the processor industry, and more and more new projects will begin to appear in this area. An example would be the gEDA tool - it allows you to design chips with an unlimited number of components. gEDA includes utilities for editing and modeling chips and tracing boards. The solution was developed for UNIX platforms, but a number of its components also work under Windows. Guidance on working with them can be found in the documentation on the project website .
Freely distributed tools provide independent organizations and startups more opportunities. It is likely that over time, new OpenROAD approaches to developing EDA tools and creating chips can become an industry standard.
What we write about in our corporate blog: