Set micro-channel news, Hon Hai Group in recent years to cultivate the Indian market, in particular Hon Hai chairman Terry Gou has repeatedly sought refuge in India, had planned in August 2015 announced plans to hit 5 billion U.S. dollars in Maharashtra in western India Maharashtra has set up a large manufacturing plant and is expected to bring about 50,000 job opportunities in the next five years. According to Taiwan media, after two years of uncertainty and negotiations, the plan has become a bad check, disappointing local Indian officials.
Indian financial media "Mint" reported that Subhash Desai, Minister of Industry of Maharashtra Government, said: "Maharashtra must skip Foxconn (Hon Hai) as early as possible," which means Hon Hai may not be able to Subhash Desai said that if Hon Hai did not invest, he would be disappointed with the media, but that did not make any difference to the state because there were other foreign direct investment plans coming in.
When the investment plan was announced more than two years ago, it was predicted that Hon Hai might help big client Apple produce iPhone in India, but later became another Wistron foundry in Taiwan, setting up a 4-inch factory near Bengaluru The iPhone SE due to the price of expensive, iPhone in the Indian market to expand slowly, an estimated market share of only 2.5%, far less than the Chinese mainland brand, the first retail store also delayed for many years has not yet opened.
Some people think that Hon Hai's successful acquisition of Japan's Sharp or affect subsequent follow-up investment plan, including non-fulfillment of investment commitments in India.Unified by Sharp into the group, Hon Hai has a strong interest in liquid crystal panels and television investment, in addition to the end of 2016 and Guangzhou Municipal government cooperation in the local investment of 9 billion US dollars to build the latest generation of LCD panel plant last year announced in the United States invested tens of billions of dollars in Wisconsin, the production of panels and television.
In February, when India's finance minister, Arun Jaitley, announced next year's budget, he said he decided to increase the import duty on smartphones and hold a positive attitude toward the industry and considered it helpful for the government to achieve the goal of "making in India." According to Moneycontrol.com Reported that the Indian government will be from the April 1 fiscal year, the import tariff on smart phones from the current 15% to 20%. In fact, some of the relevant specific parts and accessories, import duties have been previously from 7.5 ~ 10% increase to 15%.
According to another official of the industrial sector, formal negotiations between Hon Hai and the Indian government over the past two and a half years have not yielded any results. If the investment plan has been canceled and it is not wrong, including finding suitable land, labor laws, the Sino-Indian border crisis, As well as its environmental impact, etc. In fact, Tomoko, a unit of Hon Hai, already has a foundry in India to produce smartphones for customers such as Xiaomi, and a third foundry will be added in the future.
2. Intel's unique computing chip or speed up the development of quantum computers;
QuTech, the Dutch quantum computing company, announced on February 15 that it has partnered with chip maker Intel to develop a programmable dual quantum computer running on a silicon chip.
The researchers used a special type of quantum unit, the 'spin quantum unit', to run two different quantum algorithms on a silicon chip.
Spin quantum units have the advantage that they do not require harsh conditions, such as cryogenic temperatures. Essentially, spin quantum units are electrons that are activated by microwave pulses.
General quantum computer systems, such as the Intel 49-Quantum Unit Computer, rely on superconducting materials and very high temperature requirements of near zero degrees (about minus 273 degrees Celsius), making their use very small.
It is generally accepted that quantum computing should be able to do things that 'normal' computers can not do, such as simulating complex molecules or communications that can not be hacked. Quantum computing seems to change everything.
However, these are good expectations, and quantum computers, like other technologies, are subject to various factors, and quantum computers are still in their infancy.
Although people put a lot of resources, but the quantum computer can only do a few things, far from reaching the real practical level.
The industry believes that 'spin quantum units' bring hope, because it is not a real quantum computer, but with the existing common computer technology in order to achieve quantum computing.
Intel, in particular, is a global leader in silicon chip sales and can take advantage of this technology to accelerate the development of usable quantum computers.
Technicians in the white paper released state that the unique advantages of a spin-quantum unit are the electronics-level operations that work well with existing computer workstations.
The researchers also said that this new type of quantum cell system initially through experimental testing, still need to explore more performance.
At present, the development of quantum computers seems to be at a bottleneck. The technicians need the computational performance of 100 quantum units, but how to implement this technique has not been found yet.
Perhaps, Intel and QuTech silicon chip 'quantum computer' will open up a new way. 3.IoT application a variety of MCU development platform increased in importance;
The Internet of Things has brought a huge business opportunity to the MCU. As the competition between vendors is fierce, how to reduce the cost while improving the efficiency so as to achieve a better price / performance ratio has become an important topic for the industry.
Due to the diverse application needs of the Internet of Things, MCU developers have been continuously increasing their Cost-Performance Ratio and adding more new features to next-generation solutions. At the same time, Peripheral services and a good development environment to help IoT application developers to improve product performance and accelerate time to market.
Yang Zhenglian, senior product marketing manager for STMicroelectronics Asia Pacific, said (Figure 1) that in the future, consumer, commercial and industrial electronics will continue to focus on sensing and the evolution of such connectivity trends will ultimately There is a complete IoT architecture.However, currently in the evolution of this trend, so things related to the development of MCU space and application of imagination are very alarming, business opportunities can be expected.
It is also due to the many possibilities of the Internet of Things in the future, making the task of MCU more and more diverse, how to integrate the various functions between the manufacturers become an important issue.Yang Zhenglian pointed out that due to the large number of vendors into the MCU, MCU The market competition is more and more intense.In this Red Sea market, how to improve the cost performance has become an important direction of progress, various manufacturers also resorted to housekeeping skills to meet market demand.
PPAC four key points to evaluate the cost-effective MCU
In the IoT applications such as wearable devices, home automation, sensing, etc., the need to connect different functions in the same device will bring the test to the MCU. In addition to reducing hardware costs, how to improve MCU performance (Figure 2), crystal heart in assessing the effectiveness of MCU work, usually through the "PPAC" four main points to assess, which is Power (Power) Performance, Area, Code Size.
Lin Chih-ming further explained that the first in the MCU's power consumption hopes to be able to do as much as possible to save power; secondly in the computing efficiency to be high instantaneously, without delay; In addition, the relative area of the MCU must be small, at the same performance MCU or the time occupied by the smaller, can relatively increase the value; Finally, as the MCU is based on the code, we must think of ways to make it as simple as possible, which can improve MCU performance.
On the other hand, in response to the demand for IoT, Citrix convened relevant vendors of information security and IoT specifications to work together to create an IoT-specific Knect.me ecosystem whose community partners work together to provide link designers SoC development platform, software stack, application development platform and development tools to help them build highly competitive products, in response to the ever-changing Internet industry trends.
ASSP MCUs emphasize application optimization
As all things networking needs will drive a variety of different products will be equipped with connectivity, for the field of MCU will be a major challenge.Different devices in the linkage of sensing, wireless networking and MCU in series, will face different systems Integration test In this state, the ASSP MCU has become the most cost-effective option for traditional appliances when engaging in smart appliances.
Due to the diverse needs of IoT devices, Holtek has developed a variety of ASSP MCU products for different equipment needs of the Internet of Things. Taking personal health care products as an example, in recent years Blood glucose meters, sphygmomanometers, ear thermometers, and more are connected to the Internet and require the ability to run health data monitoring functions in tandem with the mobile App. However, sensing technologies vary from device to device. For an amplifier (OPA) Digital converter (A / D), input / output (I / O) and so on are all different requirements, so the MCU must also be tailored to different sensing technologies.
Cai Rongzong further said that compared with foreign manufacturers, Holtek's 2.2 billion capital is relatively small, but also relatively unaffected by the pressure of mass-produced product lines. Holtek's biggest strength is that it is flexible and flexible and efficient Production line deployment, can quickly meet the needs of a small number of diverse production, and thus to meet the needs of a variety of MCU IoT era.
On the other hand, Cai Rongzong shares, but also due to the diverse needs of the Internet of Things, the increasingly complex forms of related supply chain has also been changing.For example, the sensor vendors will also hope that the analog-digital conversion function MCU further built-in products In order to become a more complete module sales, Holtek also develops related MCU products to cooperate with this type of manufacturer, and expects to be able to create more different cooperation opportunities in response to the trend.
Therefore, in addition to thinking about the continued stability of IC, MCU strength, but also in the establishment of a subsidiary in 2016 to create science and technology, direct sales of electronic modules.Cai Rongzong shows that the product is multiplying the cluster of Holtek MCU plus PCB sales, Enable makers, students, personal studios can speed up the commercialization process, shorten the development process.If confirm the mass production, you can further shift to Holtek to buy MCU directly to reduce the cost of a large number of purchases.
Development platform to reduce MCU application development costs
The demand for IoT devices varies with usage scenarios and needs for MCUs. For a competitive MCU vendor, in addition to being competitive in cost of hardware, customer application development costs and time-to-market Must be considered an important part.Therefore, the development platform is perfect or not, but also one of the core competitiveness of MCU suppliers.
According to Jennifer Hin, director of Texas Instruments' semiconductor marketing and application embedded system (Figure 4), the characteristics of the Internet of things make the customization needs of MCUs highly valued, and the time and human resources invested by R & D are a big cost. It is not possible to react directly to the bill of materials, however, the capital invested is sometimes higher than the cost of purchasing a chip.
In the era of Internet of Things, wired communication has become an important technology that MCUs mainly must import, however, there are different connection requirements in industrial automation, building automation, smart home, etc. Taking smart home as an example, In an apartment building, Wi-Fi and BLE technology can meet most application needs, but in a smart factory context, the required connection distance may have to be met using ZigBee Mesh technology. If we consider the data to be transmitted The amount of data, many more considerations.
Jennifer CHAN believes that, MCU to achieve cost-effective, in addition to hardware, software cost considerations, reducing R & D investment in resources is also another way to improve cost-effective thinking.With the Internet of Things need to connect the MCU function, for example, how Creating a platform that creates more reusable value and delivering customized MCU products faster is also one way to create high-CP-value products.
Advanced process to help MCU performance
Due to the rise of the Internet of Things, the demand for integration and low power consumption will increase, which in turn will prompt the MCU to be more advanced. NXP Semiconductors Greater China Product Marketing Manager Microprocessor Huang Jianzhou (Figure 5), the previous MCU process is mainly 0.18 or 0.13 based, but in the future if the MCU to continue to improve the integration and reduce power consumption, it is bound to move toward 90 nm or even 40 nm process, in order to achieve the performance improvement .
Huang Jianzhou that further sharing, in order to achieve cost-effective, cost-cutting and efficiency of the two directions are indispensable.In NXP's product planning, the consumer IOT products are currently based on 32-bit MCU.However, As 8-bit MCU products still have their importance in the industrial application scenario and the market share is still high, NXP will also continue to expand and maintain the 8-bit MCU product line.
IoT pulls 32-bit demand 16-bit market is shrinking Lin Zhiming pointed out that although 8-bit and 32-bit MCU market share comparable, but the 8-bit MCU market has reached saturation, and 32-bit will be due to IoT demand continued growth, therefore, 32-bit MCU will be crystal core business in the future the main direction of promotion.On the other hand, with the artificial intelligence and increasingly complex storage needs, the future will also develop 64-bit MCU application.
Yang Zhenglian predicts that 8-bit MCUs will also gradually be replaced by 32-bit MCUs as future MCUs continue to evolve to meet the needs of the Internet of Things. The strategic layout of STMicroelectronics is based on a variety of 32-bit Versatile MCU to meet all the needs of customers.At present, STMicroelectronics has more than 700 different frequency, memory size, function of the general-purpose MCU, made more mobile App allows business people to communicate with customers to promote MCU products more smoothly Figure 6).
Under the open development environment of ARM core, R & D personnel gradually reduce the cost of developing applications by using 32-bit micro-controller (MCU), and the price of the chip also decreases due to the large number of suppliers. However, the 8-bit MCU has more Low prices, coupled with the sheer size of the low-end application market, have not surpassed 8-bit MCUs despite the catch-up of 32-bit MCU shipments. By comparison, 16-bit MCUs have become the 8-bit With 32-bit MCU MCU kill products.
Cai Rongzong that the current global market, despite the higher 32-bit MCU output value, but the sales volume is still based on 8-bit MCU bulk 16-bit MCU is in the trapped state, because the 8-bit MCU Performance continues to improve, is sufficient to meet the more low-end 16-bit MCU applications demand; 32-bit MCU costs gradually reduce, then the high-end 16-bit MCU market.
On the other hand, due to the mature market of 8-bit MCUs and 32-bits despite the backward specifications, a very friendly development environment and resources have been established under the promotion of ARM.If the manufacturer plans to invest 16 bits Market, you need to establish another set of standard language, R & D cost is too high. It is foreseeable that the future 16-bit MCU market share will gradually shrink.
Cai Rongzong pointed out that the market still can see some 16-bit MCU applications, the traditional car market will occupy a considerable proportion However, in accordance with the future of vehicles to automation, the development direction of electric vehicles, car MCU will be 16 Bit gradually transferred to 32 bits. New Electronics 4. Semiconductor manufacturing process NSD into a new selection of ADC indicators;
The recent proliferation of high-speed and very high-speed ADCs and digital processing in a variety of applications has made Oversampling a viable architectural approach to broadband and RF systems. The shrinking of semiconductor manufacturing has led to speed (Cost, power consumption, board space, etc.), allowing system designers to use a wideband converter with flat noise spectral density (NSD) Band-limited sigma-delta converters with high dynamic range required to explore a wide range of conversion and processing approaches that have changed the way designers think about signal processing and how they define the product .
The NSD and its distribution in the frequency band of interest provide in-depth insights as well as references to the converter selection process.
Noise spectral densities are often more useful than SNR specifications when comparing systems that vary widely in speed, or to explore how software-defined systems handle signals of different bandwidths. Noise spectral density does not supersede other specifications, but it is indeed a useful project worth joining in the analysis toolbox.
SNR provides full signal power information
When an SNR is included in a data converter specification it provides information about the total noise power present in all other bins and the Full-Scale signal power.
A simplified example of comparing SNR with NSD is shown in Figure 1. Assuming a clocked frequency of 75MHz for the ADC, this figure shows the Fast Fourier Transform (FFT) for output data where DC to 37.5MHz Spectrum. In this case, the signal of interest is the only strong signal that exists, which happens to be located around 2 MHz. For white noise (which usually includes quantization and thermal noise), its noise is evenly distributed over the transition On the Nyquist band, which in this case is from DC to 37.5 MHz.
Since the signal of interest is between DC and 4MHz, it is easy to first digitally post-process, filter out or discard all signals above 4MHz (leaving only the signal in the red box), where a 7/8 The noise is handled and all the signal energy is kept, essentially equal to a 9dB increase in the effective SNR. In other words, if the known signal would be in half the frequency band, we could actually remove the other half Band, but only eliminate the noise.
This leads to a useful rule of thumb that in the presence of white noise, the processing gain provides an additional 3dB / octave SNR for oversampled signals.In the example of Figure 1, this technique can be applied to more than three octaves (factor 8) to achieve a 9dB SNR improvement.
Of course, if the signal is somewhere between DC and 4MHz, there is no need to use a 75MSPS high-speed ADC to capture the signal, but only 9MSPS or 10MSPS to meet Nyquist sampling's bandwidth requirements. , It is possible to use a factor of 8 to decimate the 75 MSPS sample data to produce an effective data rate of 9.375 MSPS while retaining the Noise Floor in the band of interest.
It is important to perform the decimation correctly: simply extracting seven values from every eight samples results in a Fold Back or Alias into the band of interest, and as a result, This does not improve the signal to noise ratio, so filtering must be performed before decimation can be performed to achieve processing gain.
Even though in this case a perfect Brick Wall filter eliminates all these noise and produces the desired 3dB / octave processing gain, the fact is that filters with such characteristics do not exist.
In practice, the amount of rejection required for the stop band of the filter is a function of how much processing gain needs to be achieved, and remember that the 3dB / octave rule of thumb is based on the assumption that noise is white noise and for most But not all) this is a reasonable assumption.
However, one important exception occurs when the dynamic range is limited by the source of non-linear or other Spurious Intermodulation Products in the band. In this case, filters and discard methods may (But it may or may not) catch a spur that limits its performance and therefore require a more cautious approach to frequency planning.
SNR sampling rate is converted to noise spectral density
The situation can become more complicated when there is more than one signal in the spectrum, such as many stations on the FM radio band.
The most important factor in recovering any single signal is not the overall noise of the data converter, but the total amount of converter noise that falls into the band of interest, which requires digital filtering and post-processing to eliminate all out-of-band (Out-of-Band) noise.
There are different ways to reduce the amount of noise that falls into the red box, one of which is to choose an ADC with better SNR (smaller noise), or it can use the same SNR, but ADCs with higher clock frequencies (for example, 150MHz) can break up the spikes over a wider bandwidth, reducing the amount of noise left in the red-box region.
NSD is a better indicator
This brings with it a new question: Are there any better metrics than SNR that can be quickly compared to the converter in order to determine its performance in the red box?
This is where noise spectral densities are to be brought into the discussion. By describing the noise at spectral density (usually in dBFS / Hz over full width), it is possible to compare ADCs with different sampling rates to Determine which noise may be the least in a given application.
Table 1 shows a data converter with 70dB SNR, which illustrates the relative NSD improvement over the sampling rate range from 100MHz to 2GHz.
Table 2 compares several very different converters that use a combination of different SNRs and sample rates, but since they all have the same NSD, the total amount of noise that crosses over a 1 MHz channel is Remember, the actual resolution capability of a converter may be much larger than its effective number of bits because many converters require extra resolution to ensure that the quantization noise affects NSD Can be ignored.
In a traditional single-carrier system, it may seem absurd to use a 10GSPS converter to capture a 1MHz signal, but in a multi-carrier, software-defined environment, this may well be the exact approach a designer will take. An example is a cable set-top box that uses a 2.7GSPS to 3GSPS full spectrum tuner internally to capture a cable signal consisting of hundreds of television channels, each with a bandwidth of several MHz. For data Converter, which is traditionally expressed in dBFS / Hz as the noise spectral density, is measured in dB relative to full scale per Hz. This information provides an output of the noise level Output Referred Measurement, or in dBm / Hz or even dB mV / Hz to provide a more absolute measurement: the Input Referred Indication of the data converter noise.
SNR, full-scale voltage, input impedance, and Nyquist bandwidth can also be used to calculate the effective noise figure for the ADC, but that would be a fairly complicated calculation.
Super-sampling advantages and more
Having the ADC operate at a higher sampling rate usually means consuming more power, which includes the ADC itself and the subsequent digital processing, and the advantages of oversampling over NSD are illustrated in Table 1. But the problem persists Is super-sampling really worth it?
A better NSD is also obtained using a lower noise converter as shown in Table 2. For a system that needs to capture multiple carriers, it needs to operate at a higher sampling rate, so each carrier is However, oversampling still has some advantages.
Simplification of Antialias Filtering: The sampling behavior causes higher frequency signals (and noise) to fall back to the Nyquist band of the converter due to aliasing. Therefore, in order to avoid aliasing artifacts, these signals Must be suppressed by a filter located before the ADC, which means that the filter's Transition Band must fall between the expected FIN and the frequency of aliasing (FSAMPLE, FIN) Of the FSAMPLE, the anti-aliasing filter's transition band becomes very narrow and therefore requires a very high-order filter. Two to four times the oversampling can in principle reduce this in the analogy One limitation is the shift to work in the digital domain that is easier to handle.
The effect of a Folding Converter Distortion Product on minimizing, even for a perfect anti-aliasing filter, can lead to defects that create spikes and other distortion artifacts in the ADC, including Some very high order harmonics, which also fold the Fold Across sampling frequency, are likely to fall into the band with some limitations on the SNR within the band of interest, and at higher sampling rates, The need for the band becomes just a fraction of the Nyquist bandwidth, thereby reducing the occurrence of folding.It is worth mentioning that oversampling also helps in frequency planning to prevent other system bursts from folding back into the band.
Processing gain affects any white noise: this includes thermal and quantization noise, as well as noise from some types of jitter.
As speed-up of converters and digital processing becomes readily feasible, system designers are now increasingly using some degree of oversampling to take full advantage of these advantages, such as Noise Floor, And FFT.
Using the spectrogram to look at the depth of the noise reference for converter comparison is an attractive way, as shown in Figure 2. However, in making such comparisons, it is important to remember that the spectrum is drawn based on fast Fourier The size of the leaf transform. Larger FFTs split the bandwidth into more bins, resulting in less accumulated noise in each basket. In this case, the spectrum clearly shows Low noise floor, but this is just a graphical artifact.
In fact, the noise spectrum density has not changed (this is equivalent to a signal processing that changes the resolution bandwidth of the spectrum analyzer).
In the end, this background comparison is acceptable if the sample rate and the FFT size are the same (or scaled appropriately), but if not, the comparison can be misleading, and NSD standard is useful for direct comparison of the occasion.
So far, discussions on the handling of gain and oversampling have been based on the assumption that all the noise on the Nyquist zone of the converter is flat, which in many cases can be a reasonable approximation. There are some situations where this assumption does not hold.
For example, as mentioned earlier, while oversampling systems may offer some advantages in frequency planning and surge processing, the processing gain does not necessarily apply to the surge in addition to 1 / f noise and some Spectral Shaping is the type of oscillator phase noise, and processing gain calculations are not suitable for this situation.
One of the key circumstances in which noise is not flat occurs when a sigma-delta converter is used. The sigma-delta modulator uses feedback on the quantizer to shape the quantiser's noise and thus The noise that falls into the band of interest is reduced at the expense of out-of-band noise, as shown in Figure 3.
Even without a thorough analysis, we can see that NSD is used as a specification parameter to determine the available dynamic range in the band, especially for the sigma-delta modulator.Figure 4 shows a high speed, bandpass, Σ- A partial amplification of the noise reference for ΔADC The noise is about -160 dBFS / Hz over the entire 75 MHz band of interest (center frequency 225 MHz), providing an SNR of over 74 dBFS.
Reducing the Signal Bandwidth Reducing the Dynamic Range Improvement In Figure 5 we compared five ADCs: a 12-bit 2.5 GSPS (purple curve), a 14-bit 1.25 GSPS ADC (red curve) with 500 MSPS clocking, 1GSPS (green curve), a 14-bit 3GSPS ADC with one clock for 3GSPS (gray curve), and a different 14-bit 500MSPS (blue curve) with 500MSPS for one clock. The last is the Bandpass Σ -Δ ADC. The first five cases are evaluated using a noise floor that approximates white noise (flat) noise, while the Σ-Δ ADC has a bath-like noise spectral density with a low impurity in the band of interest News distribution, as shown in Figure 4.