Showing posts with label Low power Digital VLSI design. Show all posts
Showing posts with label Low power Digital VLSI design. Show all posts

Saturday, May 29, 2021

Low power techniques in Digital VLSI Design.

We have different low power design techniques available at the front-end and at the back-end of VLSI design flow to reduce the design's dynamic and static power. Some of the methods are clock gating, power gating, lowering supply voltage and frequency.

At lower nodes, many other techniques have been investigated and used in designs. Below we will discuss some basic and yet important techniques that are frequently used at the back-end level of  VLSI design flow.

a) Reduction in switching activity: There are many ways for reducing switching activity. We have an age-old and conventional technique known as clock gating. It is quite simple in which we block the clock signals of any block whenever the data inside the block is not changing or when the block is not required.

In the above-given figure, the Block A clock is dependent upon enable signal. Whenever the signal becomes high, it provides a clock to block A. Here we have used AND gate as a clock gating cell, but there will chances of glitches in the clock, which will be propagated to the block and inturns to functionality errors. To avoid this case, we use ICG cells (Integrated clock gating cells) available in. lib provided by the foundry based on the latch to avoid eliminate any glitch on enable signal which can be propagated through the clock.

Another technique that can help us to reduce dynamic power by lowering switching activity is replacing binary counters with a gray counter. For example: In asynchronous FIFO, we can use a gray counter instead of a binary counter to synchronize read and write pointers.

b) By reducing supply voltage: We know that the square of supply voltage is directly proportional to dynamic power. By lowering a supply voltage, we can reduce dynamic power dissipation in design. But it can impact the speed and performance of design and may give rise to subthreshold leakage currents.   

c) Multi-VT Design: If two different cell libraries of HVT and LVT are available, we can replace some parts of the design with HVT and LVT. For example, The high-performance block is synthesized by HVT cells, and the timing critical block can be synthesized by LVT cells. 

d) Multi-voltage Design:  The same IC can have blocks working on different voltages according to their performance requirements. For example, a SOC has a core and different peripherals like UART, SPI, etc., which can perform DSP algorithms. In this case, to perform such heavy processing, we need high performance from the core. We have to operate it with a high power supply, and other peripherals can operate at a lower voltage. A level shifter is used, which will shift from one voltage domain to another. Similarly, the signals leaving from one domain to another should be shifted using level shifters.

In the above figure, the level shifter is connected with both voltage supplies and shifting the signal levels from low voltage domain to high voltage domain. 

e) Power Gating: It is a technique in which we can turn off the power supply or a part of IC by using a power gating cell. Thus it can help in reducing the dynamic as well as static power of the design. This technique can be implemented using a power controller, power network, isolation cells, and retention register. A power controller is a logic block that will signal a specific block to turn on or turn off the power supply.
The power gating cell is implemented using an HVT transistor from MTCMOS (multi-threshold CMOS) technology to minimize leakage currents and static power.


In the above figure, we have a switch connected in between VDD and block is known as a header switch, and if a switch is connected in between the block and VSS, it is known as a Footer switch. This is strategy is also known as a coarse-grain strategy as power gating is directly implemented over a block.


                a) Drain header and power footer gating.     b)Drain footer and power header gating

In the above-given figure, we have some advanced power gating and drain gating techniques. This is strategy is also known as a fine-grain strategy.
 

Fig. Isolation cells with an always-on block and power down block

Isolation cells: They are used with always-on block and power down the block, where signal leaving from power down the block is connected with isolation cell to provide a signal to always-on block. When power is up, it directly provides a signal coming from the block to an always-on block. If the block is turned off, the isolation cell provides logic  0 or 1 rather than any unknown value to avoid leakage currents in an always-on block. 

Retention register: The data residing inside the power down the block can be saved by storing the data inside the retention register in power-down mode. Whenever the block is powered up, the same data can be restored from retention registers. The retention register should always be powered up.

f) Dynamic voltage and frequency scaling: It is the technique where we change the frequency and voltage of the designs according to workload upon them; if low performance is required, we can use low voltage and low frequency. For high performance, the same block should be connected to higher voltage levels. To achieve this, we need to provide different voltage levels inside the chip to switch in between them whenever it's required. We can extend this by using power gating to shut down the specific block when it is not in use.

In the next article, we will discuss UPF and how we can specify the information related to supply network, isolation cells, retention registers, etc., in our design for power management.

Thank you for reading this article.

Feel free to reach me at:
E-mail id: hemantjuneja710@gmail.com 
LinkedIn: https://www.linkedin.com/in/hemant-juneja-84aa6313b/ 

Please comment down your suggestions, Happy learning :)


References:
1. https://lmr.fi/int/multi-voltage-design-power-management/
2. Panwar, Shikha & Piske, Mayuresh & Madgula, Aatreya. (2014). Performance Analysis of Modified Drain Gating Techniques for Low Power and High-Speed Arithmetic Circuits. VLSI Design. 2014. 1-5. 10.1155/2014/380362. 
3. T. Hattori, "Design methodology of low-power microprocessors," Proceedings of the ASP-DAC Asia and South Pacific Design Automation Conference, 2003., Kitakyushu, Japan, 2003, pp. 390-393, DOI: 10.1109/ASPDAC.2003.1195046
4. https://semiengineering.com/knowledge_centers/low-power/techniques/power-isolation/


Tuesday, May 25, 2021

Power consumption in ASIC.

In older designs, we were worried about timing and area to save the cost of the design but now the power is becoming crucial day by day due to development in the manufacturing unit and leading to the scaling of technologies from ยตm to nm and impacted the density of transistors to increase in the same area causing more power consumption and has become a worrisome topic. Increased power consumption in design leads to an added cost of manufacturing the IC due to the usage of ceramic packaging of IC, modified heat sinks, and the cooling mechanism that will be needed to save IC from malfunctioning.

Companies are investing a lot of money and resources in R&D to solve this issue. Each company has a specific PPA team that works on trade-offs between power, performance, and area.

Power consumption is broadly divided into two types,

 1. Dynamic power: This type of power consumption is due to switching activity in design and internal power consumption. Switching power involves charging and discharging of the external capacitances in design running on Vdd and frequency.


P(Dynamic) = (Vdd)^2*C*F 
Vdd: Power supply, C: Load capacitance, F: Frequency. 

Internal power consumption occurs for a very short duration of time due to a short circuit current, in which our PMOS and NMOS are directly connected to each other and form a conducting path between Vdd and ground. For example in an inverter as per DC characteristic, there is a point in VTC curve at which vin=vout which is also known as switching threshold in which pmos and nmos are in a saturation region.




2. Leakage Power: Leakage power was negligible in older technologies but at deep submicron technologies it has been increased, sometimes it is even comparable to the dynamic power consumption of the design. The subthreshold current, gate current and reverse bias PN junction current contribute to the leakage power. We will discuss each of them one by one:
a) Subthreshold current: Due to scaling, we have reduced gate threshold voltage of mosfet's because of which MOSFET can conduct in small amount even when the gate voltage is less than the threshold voltage and current can flow from drain to source which is also known as sub-threshold current and contributes to power leakage.

b) Reverse bias PN junction current:  From the basics of the PN junction, we know it can conduct current in forward bias condition, but in reverse bias junction due to large depletion width it can't conduct any current. But there is a small amount of current which flows in reverse bias condition due to the minority charge carrier which is known as reverse saturation current. The same current flows here in the PN junction formed by drain and body/substrate in reverse bias condition. 

c) Gate leakage current: It is because of the thin oxide layer, which is too small and has only a few layers of oxide atoms because of which tunnelling effect can occur and current can flow from gate to source or drain.

The power leakage occurs mainly in an idle. But it can be reduced to a greater extent by using HVT cells and lowering voltage supply or by turning off design when not in use by using power and sleep cells.

We will discuss different types of techniques that can be used and how we can implement them in design in the next article.

Thank you for reading this article.

Feel free to reach me at:
E-mail id: hemantjuneja710@gmail.com 
LinkedIn: https://www.linkedin.com/in/hemant-juneja-84aa6313b/ 

Please comment down your suggestions below, Happy learning :)