| jupytext | kernelspec | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
|
The number of transistors reported per a given chip plotted on a log scale in the y axis with the date of introduction on the linear scale x-axis. The blue data points are from a transistor count table. The red line is an ordinary least squares prediction and the orange line is Moore's law.
In 1965, engineer Gordon Moore predicted that transistors on a chip would double every two years in the coming decade [1, 2]. You'll compare Moore's prediction against actual transistor counts in the 53 years following his prediction. You will determine the best-fit constants to describe the exponential growth of transistors on semiconductors compared to Moore's Law.
- Load data from a *.csv file
- Perform linear regression and predict exponential growth using ordinary least squares
- You'll compare exponential growth constants between models
- Share your analysis in a file:
- as NumPy zipped files
*.npz - as a
*.csvfile
- as NumPy zipped files
- Assess the amazing progress semiconductor manufacturers have made in the last five decades
1. These packages:
- NumPy
- Matplotlib
- statsmodels ordinary linear regression
imported with the following commands
import matplotlib.pyplot as plt
import numpy as np
import statsmodels.api as sm
2. Since this is an exponential growth law you need a little background in doing math with natural logs and exponentials.
You'll use these NumPy, Matplotlib, and statsmodels functions:
-
np.loadtxt: this function loads text into a NumPy array -
np.log: this function takes the natural log of all elements in a NumPy array -
np.exp: this function takes the exponential of all elements in a NumPy array -
lambda: this is a minimal function definition for creating a function model -
plt.semilogy: this function will plot x-y data onto a figure with a linear x-axis and$\log_{10}$ y-axisplt.plot: this function will plot x-y data on linear axes -
sm.OLS: find fitting parameters and standard errors using the statsmodels ordinary least squares model - slicing arrays: view parts of the data loaded into the workspace, slice the arrays e.g.
x[:10]for the first 10 values in the array,x - boolean array indexing: to view parts of the data that match a given condition use boolean operations to index an array
-
np.block: to combine arrays into 2D arrays -
np.newaxis: to change a 1D vector to a row or column vector -
np.savezandnp.savetxt: these two functions will save your arrays in zipped array format and text, respectively
+++
Your empirical model assumes that the number of transistors per semiconductor follows an exponential growth,
where
You determine these constants for Moore's law by specifying the rate for added transistors, 2, and giving an initial number of transistors for a given year.
You state Moore's law in an exponential form as follows,
Where
-
$\dfrac{\text{transistor_count}(\text{year} +2)}{\text{transistor_count}(\text{year})} = 2 = \dfrac{e^{B_M}e^{A_M \text{year} + 2A_M}}{e^{B_M}e^{A_M \text{year}}} = e^{2A_M} \rightarrow A_M = \frac{\log(2)}{2}$ -
$\log(2250) = \frac{\log(2)}{2}\cdot 1971 + B_M \rightarrow B_M = \log(2250)-\frac{\log(2)}{2}\cdot 1971$
so Moore's law stated as an exponential function is
where
Since the function represents Moore's law, define it as a Python
function using
lambda
A_M = np.log(2) / 2
B_M = np.log(2250) - A_M * 1971
Moores_law = lambda year: np.exp(B_M) * np.exp(A_M * year)
In 1971, there were 2250 transistors on the Intel 4004 chip. Use
Moores_law to check the number of semiconductors Gordon Moore would expect
in 1973.
ML_1971 = Moores_law(1971)
ML_1973 = Moores_law(1973)
print("In 1973, G. Moore expects {:.0f} transistors on Intels chips".format(ML_1973))
print("This is x{:.2f} more transistors than 1971".format(ML_1973 / ML_1971))
Now, make a prediction based upon the historical data for
semiconductors per chip. The Transistor Count
[4]
each year is in the transistor_data.csv file. Before loading a *.csv
file into a NumPy array, its a good idea to inspect the structure of the
file first. Then, locate the columns of interest and save them to a
variable. Save two columns of the file to the array, data.
Here, print out the first 10 rows of transistor_data.csv. The columns are
| Processor | MOS transistor count | Date of Introduction | Designer | MOSprocess | Area |
|---|---|---|---|---|---|
| Intel 4004 (4-bit 16-pin) | 2250 | 1971 | Intel | "10,000 nm" | 12 mm² |
| ... | ... | ... | ... | ... | ... |
! head transistor_data.csv
You don't need the columns that specify Processor, Designer, MOSprocess, or Area. That leaves the second and third columns, MOS transistor count and Date of Introduction, respectively.
Next, you load these two columns into a NumPy array using
np.loadtxt.
The extra options below will put the data in the desired format:
delimiter = ',': specify delimeter as a comma ',' (this is the default behavior)usecols = [1,2]: import the second and third columns from the csvskiprows = 1: do not use the first row, because its a header row
data = np.loadtxt("transistor_data.csv", delimiter=",", usecols=[1, 2], skiprows=1)
You loaded the entire history of semiconducting into a NumPy array named
data. The first column is the MOS transistor count and the second
column is the Date of Introduction in a four-digit year.
Next, make the data easier to read and manage by assigning the two
columns to variables, year and transistor_count. Print out the first
10 values by slicing the year and transistor_count arrays with
[:10]. Print these values out to check that you have the saved the
data to the correct variables.
year = data[:, 1] # grab the second column and assign
transistor_count = data[:, 0] # grab the first column and assign
print("year:\t\t", year[:10])
print("trans. cnt:\t", transistor_count[:10])
You are creating a function that predicts the transistor count given a
year. You have an independent variable, year, and a dependent
variable, transistor_count. Transform the independent variable to
log-scale,
transistor_count[i]
resulting in a linear equation,
yi = np.log(transistor_count)
Your model assume that yi is a function of year. Now, find the best-fit model that minimizes the difference between
This sum of squares error can be succinctly represented as arrays as such
where
year[:,np.newaxis]: takes the 1D array with shape(179,)and turns it into a 2D column vector with shape(179,1)**[1, 0]: stacks two columns, in the first column isyear**1and the second column isyear**0 == 1
Z = year[:, np.newaxis] ** [1, 0]
Now that you have the created a matrix of regressors, sm.OLS.
model = sm.OLS(yi, Z)
Now, you can view the fitting constants, fit and print the
summary to view results as such,
results = model.fit()
print(results.summary())
The OLS Regression Results summary gives a lot of information about
the regressors,
=================================
coef std err
---------------------------------
x1 0.3416 0.006
const -666.3264 11.890
=================================
where x1 is slope, const is the intercept,
std error gives the precision of constants
AB with
results.params and assign x1 and constant.
AB = results.params
A = AB[0]
B = AB[1]
Did manufacturers double the transistor count every two years? You have the final formula,
where increase in number of transistors is
print("Rate of semiconductors added on a chip every 2 years:")
print(
"\tx{:.2f} +/- {:.2f} semiconductors per chip".format(
np.exp((A) * 2), 2 * A * np.exp(2 * A) * 0.006
)
)
Based upon your least-squares regression model, the number of
semiconductors per chip increased by a factor of
Here, use
plt.semilogy
to plot the number of transistors on a log-scale and the year on a
linear scale. You have defined a three arrays to get to a final model
and
your variables, transistor_count, year, and yi all have the same
dimensions, (179,). NumPy arrays need the same dimensions to make a
plot. The predicted number of transistors is now
+++
In the next plot, use the
fivethirtyeight
style sheet.
The style sheet replicates
https://fivethirtyeight.com elements. Change the matplotlib style with
plt.style.use.
transistor_count_predicted = np.exp(B) * np.exp(A * year)
transistor_Moores_law = Moores_law(year)
plt.style.use("fivethirtyeight")
plt.semilogy(year, transistor_count, "s", label="MOS transistor count")
plt.semilogy(year, transistor_count_predicted, label="linear regression")
plt.plot(year, transistor_Moores_law, label="Moore's Law")
plt.title(
"MOS transistor count per microprocessor\n"
+ "every two years \n"
+ "Transistor count was x{:.2f} higher".format(np.exp(A * 2))
)
plt.xlabel("year introduced")
plt.legend(loc="center left", bbox_to_anchor=(1, 0.5))
plt.ylabel("# of transistors\nper microprocessor")
A scatter plot of MOS transistor count per microprocessor every two years with a red line for the ordinary least squares prediction and an orange line for Moore's law.
The linear regression captures the increase in the number of transistors per semiconductors each year. In 2015, semiconductor manufacturers claimed they could not keep up with Moore's law anymore. Your analysis shows that since 1971, the average increase in transistor count was x1.98 every 2 years, but Gordon Moore predicted it would be x2 every 2 years. That is an amazing prediction.
Consider the year 2017. Compare the data to your linear regression model and Gordon Moore's prediction. First, get the transistor counts from the year 2017. You can do this with a Boolean comparator,
year == 2017.
Then, make a prediction for 2017 with Moores_law defined above
and plugging in your best fit constants into your function
A great way to compare these measurements is to compare your prediction
and Moore's prediction to the average transistor count and look at the
range of reported values for that year. Use the
plt.plot
option,
alpha=0.2,
to increase the transparency of the data. The more opaque the points
appear, the more reported values lie on that measurement. The green
transistor_count2017 = transistor_count[year == 2017]
print(
transistor_count2017.max(), transistor_count2017.min(), transistor_count2017.mean()
)
y = np.linspace(2016.5, 2017.5)
your_model2017 = np.exp(B) * np.exp(A * y)
Moore_Model2017 = Moores_law(y)
plt.plot(
2017 * np.ones(np.sum(year == 2017)),
transistor_count2017,
"ro",
label="2017",
alpha=0.2,
)
plt.plot(2017, transistor_count2017.mean(), "g+", markersize=20, mew=6)
plt.plot(y, your_model2017, label="Your prediction")
plt.plot(y, Moore_Model2017, label="Moores law")
plt.ylabel("# of transistors\nper microprocessor")
plt.legend()
The result is that your model is close to the mean, but Gordon Moore's prediction is closer to the maximum number of transistors per microprocessor produced in 2017. Even though semiconductor manufacturers thought that the growth would slow, once in 1975 and now again approaching 2025, manufacturers are still producing semiconductors every 2 years that nearly double the number of transistors.
The linear regression model is much better at predicting the
average than extreme values because it satisfies the condition to
minimize
+++
The last step, is to share your findings. You created
new arrays that represent a linear regression model and Gordon Moore's
prediction. You started this process by importing a csv file into a NumPy
array using np.loadtxt, to save your model use two approaches
np.savez: save NumPy arrays for other Python sessionsnp.savetxt: save a csv file with the original data and your predicted data
Using np.savez, you can save thousands of arrays and give them names. The
function np.load will load the arrays back into the workspace as a
dictionary. You'll save a five arrays so the next user will have the year,
transistor count, predicted transistor count, Gordon Moore's
predicted count, and fitting constants. Add one more variable that other users can use to
understand the model, notes.
notes = "the arrays in this file are the result of a linear regression model\n"
notes += "the arrays include\nyear: year of manufacture\n"
notes += "transistor_count: number of transistors reported by manufacturers in a given year\n"
notes += "transistor_count_predicted: linear regression model = exp({:.2f})*exp({:.2f}*year)\n".format(
B, A
)
notes += "transistor_Moores_law: Moores law =exp({:.2f})*exp({:.2f}*year)\n".format(
B_M, A_M
)
notes += "regression_csts: linear regression constants A and B for log(transistor_count)=A*year+B"
print(notes)
np.savez(
"mooreslaw_regression.npz",
notes=notes,
year=year,
transistor_count=transistor_count,
transistor_count_predicted=transistor_count_predicted,
transistor_Moores_law=transistor_Moores_law,
regression_csts=AB,
)
results = np.load("mooreslaw_regression.npz")
print(results["regression_csts"][1])
! ls
The benefit of np.savez is you can save hundreds of arrays with
different shapes and types. Here, you saved 4 arrays that are double
precision floating point numbers shape = (179,), one array that was
text, and one array of double precision floating point numbers shape =
(2,). This is the preferred method for saving NumPy arrays for use in
another analysis.
If you want to share data and view the results in a table, then you have to
create a text file. Save the data using np.savetxt. This
function is more limited than np.savez. Delimited files, like csv's,
need 2D arrays.
Prepare the data for export by creating a new 2D array whose columns contain the data of interest.
Use the header option to describe the data and the columns of
the file. Define another variable that contains file
information as head.
head = "the columns in this file are the result of a linear regression model\n"
head += "the columns include\nyear: year of manufacture\n"
head += "transistor_count: number of transistors reported by manufacturers in a given year\n"
head += "transistor_count_predicted: linear regression model = exp({:.2f})*exp({:.2f}*year)\n".format(
B, A
)
head += "transistor_Moores_law: Moores law =exp({:.2f})*exp({:.2f}*year)\n".format(
B_M, A_M
)
head += "year:, transistor_count:, transistor_count_predicted:, transistor_Moores_law:"
print(head)
Build a single 2D array to export to csv. Tabular data is inherently two
dimensional. You need to organize your data to fit this 2D structure.
Use year, transistor_count, transistor_count_predicted, and
transistor_Moores_law as the first through fourth columns,
respectively. Put the calculated constants in the header since they do
not fit the (179,) shape. The
np.block
function appends arrays together to create a new, larger array. Arrange
the 1D vectors as columns using
np.newaxis
e.g.
>>> year.shape
(179,)
>>> year[:,np.newaxis].shape
(179,1)output = np.block(
[
year[:, np.newaxis],
transistor_count[:, np.newaxis],
transistor_count_predicted[:, np.newaxis],
transistor_Moores_law[:, np.newaxis],
]
)
Creating the mooreslaw_regression.csv with np.savetxt, use three
options to create the desired file format:
X = output: useoutputblock to write the data into the filedelimiter = ',': use commas to separate columns in the fileheader = head: use the headerheaddefined above
np.savetxt("mooreslaw_regression.csv", X=output, delimiter=",", header=head)
! head mooreslaw_regression.csv
In conclusion, you have compared historical data for semiconductor
manufacturers to Moore's law and created a linear regression model to
find the average number of transistors added to each microprocessor
every two years. Gordon Moore predicted the number of transistors would
double every two years from 1965 through 1975, but the average growth
has maintained a consistent increase of mooreslaw_regression.npz, or as another csv,
mooreslaw_regression.csv. The amazing progress in semiconductor
manufacturing has enabled new industries and computational power. This
analysis should give you a small insight into how incredible this growth
has been over the last half-century.
+++
- "Moore's Law." Wikipedia article. Accessed Oct. 1, 2020.
- Moore, Gordon E. (1965-04-19). "Cramming more components onto integrated circuits". intel.com. Electronics Magazine. Retrieved April 1, 2020.
- Courtland, Rachel. "Gordon Moore: The Man Whose Name Means Progress." IEEE Spectrum. 30 Mar. 2015..
- "Transistor Count." Wikipedia article. Accessed Oct. 1, 2020.