Design Article
Comment
Bhola_#1
Good point. Another thing to notice is Absolute max rating on datasheets. Many ...
nGENEr
Yes, you make good points also. There are a few advantages to a CMOS part. ...
Go beyond the datasheet, Part 2: Understand the considerations
Ivan Garcia, Gene Frantz
10/11/2010 9:44 AM EDT
Clock speed trends
One popular parameter that often needs to reach its limit is the operating processor speed. More MHz equals more processing power (both performance and power dissipation). More processing power yields higher resolutions, more channels and fewer devices on a board. Given that silicon manufacturers need to productize a particular device for a wide audience, they need to test across a rigorous array of specifications; each device shipped to a customer must meet all of the data sheet specifications. Moreover, in order for a device to be successful in the market, it needs to be attractive to multiple applications, each with its own set of requirements. When putting together this device, it then needs to meet and exceed each and every one of these multi-application requirements.
That begs the question: What if this device had to meet your application requirements only?
Figure 1 shows the behavior of an internal memory test for a DSP at various voltages running at a high temperature. Figure 2 shows the same test, on the same DSP at room temperature.
As you can see, if this device is rated to run for a case temperature as high as 85ºC at, say, 1.2 V, the manufacturer must rate this device to run at 360 MHz at most.
However, if an application is designed such that the device will never operate above 25ºC (or less than 85ºC), you could theoretically run the same device at almost 500 MHz for this particular application.
Fig. 1: Voltage vs. MHz for a DSP at high temperature (85ºC)
Fig. 2: Voltage vs. MHz for same DSP at 25ºC (room temperature)
Another way to look at this data is that if the application environment called for an 85ºC operating temperature then increasing the voltage from 1.2 V to 1.4 V means this device could run at a speed higher than 700 MHz.
Lastly, if power is a consideration, one could run this application at 25ºC while dropping the voltage and keeping the clock speed constant in order to decrease the operating voltage.
Note that this data is for a specific test condition only. The important thing to remember here is that a particular application may only need a subset of the test conditions for a given device, hence giving an amount of flexibility with these parameters.
Next Page: Tweaking the performance, power budget
One popular parameter that often needs to reach its limit is the operating processor speed. More MHz equals more processing power (both performance and power dissipation). More processing power yields higher resolutions, more channels and fewer devices on a board. Given that silicon manufacturers need to productize a particular device for a wide audience, they need to test across a rigorous array of specifications; each device shipped to a customer must meet all of the data sheet specifications. Moreover, in order for a device to be successful in the market, it needs to be attractive to multiple applications, each with its own set of requirements. When putting together this device, it then needs to meet and exceed each and every one of these multi-application requirements.
That begs the question: What if this device had to meet your application requirements only?
Figure 1 shows the behavior of an internal memory test for a DSP at various voltages running at a high temperature. Figure 2 shows the same test, on the same DSP at room temperature.
As you can see, if this device is rated to run for a case temperature as high as 85ºC at, say, 1.2 V, the manufacturer must rate this device to run at 360 MHz at most.
However, if an application is designed such that the device will never operate above 25ºC (or less than 85ºC), you could theoretically run the same device at almost 500 MHz for this particular application.
Fig. 1: Voltage vs. MHz for a DSP at high temperature (85ºC)
Fig. 2: Voltage vs. MHz for same DSP at 25ºC (room temperature)
Another way to look at this data is that if the application environment called for an 85ºC operating temperature then increasing the voltage from 1.2 V to 1.4 V means this device could run at a speed higher than 700 MHz.
Lastly, if power is a consideration, one could run this application at 25ºC while dropping the voltage and keeping the clock speed constant in order to decrease the operating voltage.
Note that this data is for a specific test condition only. The important thing to remember here is that a particular application may only need a subset of the test conditions for a given device, hence giving an amount of flexibility with these parameters.
Next Page: Tweaking the performance, power budget
|
|
|
|
|
Navigate to related information
patrick.mannion
10/11/2010 12:18 PM EDT
Here in Part 2, Gene and Ivan elaborate on what you need to consider when going beyond the datasheet. Kudos to them for putting the guidelines together. What's been your experience of 'going beyond'?
Sign in to Reply
hm
10/12/2010 5:53 AM EDT
Some time we have encountered and employed hidden features of IC for test and debug purpose. However, going beyond datasheet is very high risk and it is very difficult to get approval from manufacturer or you supervisor. I may like to mention one example below. I was designing Hotlink for US DoD project. The earlier version was 160Mbps, Hotlink I working on 5Vdc. In new design I was employing Xilinx FPGA and main working voltage was 3.3Vdc. I wanted to employ Hotlink II IC which was 3.3Vdc and had many other advantages. However, Hotlink II was not characterised to work at 160 mbps. I wrote to Cypress Semiconductor for help. But since our temperature range was -55 C to 71 C and it was military application, they never were able to say yes to this application even though IC did worked for 160 mbps. I talked to other designers in our team but to no avail. Eventually, I had to use only Hotlink I IC.
Sign in to Reply
nGENEr
10/16/2010 4:25 PM EDT
You are correct. That is why is says in the introductin that using the device outside of the data sheet voids the warrantee.
But, depending on the application, how far outside of the spec you want to run and which parameters you are exceeding, there may never be an issue. For example, driving the clock below the minimum of the data sheet should always be possible. Just the same clocking above the maximum of the data sheet should also be fine given you understand that you may be reducing the life of the product, or you may need to reduce the temperature range of operation. Remember we're aren't suggesting using a new feature, but just simple extensions to the specification.
For a military product, I would never do this unless I was prepared to either guarantee the device my self or work with the vendor to give me a special part number for my application (along with the special part number, a data sheet and warrantee for the new spec.
You make one other good point. Once you go beyond the data sheet, our applications engineers can't support you. But, once again this may not be a huge issue to you. And, if it is, stick to the data sheet.
Sign in to Reply
Sensorguy_0623
10/15/2010 12:00 PM EDT
Very enlightening information on the ability to increase clock speeds. One level of concern that I would have about depending on the going beyond specs, is what happens in the case of a die change. You could be in a position where the product works fine with lots of margin until a die shrink or similar change happens. Now, you need to ship product and can't get the old parts. I took over an ECL base product years ago where a needle generator was designed in. The product had been in production for 3 years. One day, everything in production was failing. The only fix available was to patch in a short piece of coax to obtain the required delay. You have some very good points, but go in with your eyes wide open and make sure that you have margin.
Sign in to Reply
nGENEr
10/16/2010 4:34 PM EDT
Yes, you make good points also. There are a few advantages to a CMOS part. First, each new technology node is typically faster, so exceeding the clock speed should be transparant to a new die. But, then there is always the issue of fixing a bug and having it affect you. I've had that happen to me on occasion. But generally was an easy fix on my board. Second, we generally don't take an old device to a new technology node and keep the same part number. Lastly, your product life cycle is probably much shorter than our life cycle. And, if it isn't, then it may not be wise to use it out of spec.
One additional point. A good way of using the information in the paper is for creating prototype systems and early production when the device isn't quite good enough for the application. But, with time you know you can better optimize the code and get it within spec. This gives you a short term path until the real optimized solution can be realized.
Hope this makes sense.
Sign in to Reply
Bhola_#1
10/30/2010 1:05 AM EDT
Good point. Another thing to notice is Absolute max rating on datasheets. Many a times, we misunderstand this section. Max rating doesnot guarantee performance for the device but tells that device may be damaged if exceeded this value.
Sign in to Reply